diff --git a/configs/config_1.14G_dp16_tp8_pp1_acc2_mbs16_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp16_tp8_pp1_acc2_mbs16_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..71b17183526d58fd155decf4e583a4af5a1013b1 --- /dev/null +++ b/configs/config_1.14G_dp16_tp8_pp1_acc2_mbs16_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp16_tp8_pp1_acc2_mbs16_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp1_tp1_pp8_acc64_mbs1_seq2048_zero0_tpmodeRED_l16_h2048_heads32.yaml b/configs/config_1.14G_dp1_tp1_pp8_acc64_mbs1_seq2048_zero0_tpmodeRED_l16_h2048_heads32.yaml new file mode 100644 index 0000000000000000000000000000000000000000..11e253981e6eacecc5d32e87d832cc05ffb57af3 --- /dev/null +++ b/configs/config_1.14G_dp1_tp1_pp8_acc64_mbs1_seq2048_zero0_tpmodeRED_l16_h2048_heads32.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp1_tp1_pp8_acc64_mbs1_seq2048_zero0_tpmodeRED_l16_h2048_heads32 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp1_pp4_acc2_mbs32_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp1_pp4_acc2_mbs32_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e348570872cf5fb410f1b9a5de6c377f57e9536a --- /dev/null +++ b/configs/config_1.14G_dp2_tp1_pp4_acc2_mbs32_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp1_pp4_acc2_mbs32_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp4_pp1_acc128_mbs2_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp4_pp1_acc128_mbs2_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..90655b16d8dda9424d1864289fb6b17a9542dc72 --- /dev/null +++ b/configs/config_1.14G_dp2_tp4_pp1_acc128_mbs2_seq8192_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp4_pp1_acc128_mbs2_seq8192_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp4_pp1_acc4_mbs16_seq32768_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp4_pp1_acc4_mbs16_seq32768_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8f6efb2c7858d0b43158cdc078205a2c37f5d23b --- /dev/null +++ b/configs/config_1.14G_dp2_tp4_pp1_acc4_mbs16_seq32768_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp4_pp1_acc4_mbs16_seq32768_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp4_pp1_acc4_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp4_pp1_acc4_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..89c2fded5c14392a55c8dbb57b2062161a56916c --- /dev/null +++ b/configs/config_1.14G_dp2_tp4_pp1_acc4_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp4_pp1_acc4_mbs16_seq8192_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp64_pp1_acc2_mbs32_seq32768_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp64_pp1_acc2_mbs32_seq32768_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0c86b9b564411dbbbbd0a2cf8f8820cd8785fe40 --- /dev/null +++ b/configs/config_1.14G_dp2_tp64_pp1_acc2_mbs32_seq32768_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp64_pp1_acc2_mbs32_seq32768_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp8_pp1_acc256_mbs4_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp8_pp1_acc256_mbs4_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dcaae981a38f0b4338ffc13d313867fa0ddb0bcc --- /dev/null +++ b/configs/config_1.14G_dp2_tp8_pp1_acc256_mbs4_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp8_pp1_acc256_mbs4_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 256 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp32_tp16_pp1_acc1_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp32_tp16_pp1_acc1_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..887a8a790e4e2080cb185ed0dcf9d8d6a5b69061 --- /dev/null +++ b/configs/config_1.14G_dp32_tp16_pp1_acc1_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp32_tp16_pp1_acc1_mbs4_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp32_tp8_pp1_acc1_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp32_tp8_pp1_acc1_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..86c07d9f05a691b2d234996e6c3e75f18b330503 --- /dev/null +++ b/configs/config_1.14G_dp32_tp8_pp1_acc1_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp32_tp8_pp1_acc1_mbs1_seq32768_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp64_pp1_acc64_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp64_pp1_acc64_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b13ab4aa2150082282fb150b08193d7df98b1af2 --- /dev/null +++ b/configs/config_1.14G_dp4_tp64_pp1_acc64_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp64_pp1_acc64_mbs2_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp8_pp1_acc2_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp8_pp1_acc2_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a74d9211a49fbaee4d7d7dacade7fe2bcb0be2d0 --- /dev/null +++ b/configs/config_1.14G_dp4_tp8_pp1_acc2_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp8_pp1_acc2_mbs4_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp16_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp8_tp16_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6d7cd074c8e49ab834e2139a6ccbcb76af410f09 --- /dev/null +++ b/configs/config_1.14G_dp8_tp16_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp16_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp16_pp1_acc4_mbs4_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp8_tp16_pp1_acc4_mbs4_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0735283a29e693f11eee76d37edaa1d3a47b5622 --- /dev/null +++ b/configs/config_1.14G_dp8_tp16_pp1_acc4_mbs4_seq8192_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp16_pp1_acc4_mbs4_seq8192_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp32_pp1_acc4_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp8_tp32_pp1_acc4_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eb0420368e3394d3a4f92b3ce1e526cc9678c090 --- /dev/null +++ b/configs/config_1.14G_dp8_tp32_pp1_acc4_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp32_pp1_acc4_mbs1_seq32768_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp32_pp1_acc8_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp8_tp32_pp1_acc8_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f2ae6d4bb3a9f534e6d7f4bae678d98fa2783207 --- /dev/null +++ b/configs/config_1.14G_dp8_tp32_pp1_acc8_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp32_pp1_acc8_mbs2_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp128_tp1_pp2_acc16_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp128_tp1_pp2_acc16_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..181cb39464523e0c9db48bb1d19424830915cd11 --- /dev/null +++ b/configs/config_1.34G_dp128_tp1_pp2_acc16_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp128_tp1_pp2_acc16_mbs1_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dfc178847076f3c6b1a96fbbe637bb9a07c65444 --- /dev/null +++ b/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp128_tp4_pp1_acc1_mbs4_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp8_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp8_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fb9c5cec039fe8a7bda10bb17a71a6fb2b79bf22 --- /dev/null +++ b/configs/config_1.34G_dp16_tp8_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp8_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp8_pp2_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp8_pp2_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0b0542927cdfb014430866c5677af01ff47ea995 --- /dev/null +++ b/configs/config_1.34G_dp1_tp8_pp2_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp1_tp8_pp2_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp8_pp2_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp8_pp2_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eebbe5e24c12a33e0a2ca417dfc0635a80fb606a --- /dev/null +++ b/configs/config_1.34G_dp1_tp8_pp2_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp1_tp8_pp2_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp16_pp1_acc2_mbs128_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp16_pp1_acc2_mbs128_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..601cbe2b395ed059696f9b6392773710f3fff7c3 --- /dev/null +++ b/configs/config_1.34G_dp2_tp16_pp1_acc2_mbs128_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp16_pp1_acc2_mbs128_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp16_pp4_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp16_pp4_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..44f0679fd77ffdf924a43f2ddb910b59ac6f35f7 --- /dev/null +++ b/configs/config_1.34G_dp2_tp16_pp4_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp16_pp4_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp1_pp4_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp1_pp4_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..68218789abcc53d1d7bd56d8b78a7bcfac3cf228 --- /dev/null +++ b/configs/config_1.34G_dp2_tp1_pp4_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp1_pp4_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp256_pp1_acc32_mbs2_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp2_tp256_pp1_acc32_mbs2_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0f4d0b307d0a252a60e2c6e7262af4bb53f5cfcc --- /dev/null +++ b/configs/config_1.34G_dp2_tp256_pp1_acc32_mbs2_seq8192_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp256_pp1_acc32_mbs2_seq8192_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp4_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp32_tp4_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..283fa5ece84ab5a7cb64f2124c30323dc43295d9 --- /dev/null +++ b/configs/config_1.34G_dp32_tp4_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp4_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp4_pp1_acc32_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp32_tp4_pp1_acc32_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..598be62d3e8d230a7976daf5f667ad7640dcf8e0 --- /dev/null +++ b/configs/config_1.34G_dp32_tp4_pp1_acc32_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp4_pp1_acc32_mbs2_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp16_pp1_acc2_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp16_pp1_acc2_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eb9bbfe22010af90163059f8ba768827634cb75a --- /dev/null +++ b/configs/config_1.34G_dp4_tp16_pp1_acc2_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp16_pp1_acc2_mbs16_seq32768_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp16_pp1_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp16_pp1_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f5351cd9578d52f2afbd5b859e81bbc4175e45ed --- /dev/null +++ b/configs/config_1.34G_dp4_tp16_pp1_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp16_pp1_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp2_pp2_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp2_pp2_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3f8beae91e1f3af10f43892a0e06bdd923e47de8 --- /dev/null +++ b/configs/config_1.34G_dp4_tp2_pp2_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp2_pp2_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp32_pp1_acc1_mbs8_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp32_pp1_acc1_mbs8_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..727248068c0372ef3d5061cf26bbcf4db8c0e9c1 --- /dev/null +++ b/configs/config_1.34G_dp4_tp32_pp1_acc1_mbs8_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp32_pp1_acc1_mbs8_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp32_pp1_acc32_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp32_pp1_acc32_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bc5b41471f77dc961fabd88e4b867fde7d444ff2 --- /dev/null +++ b/configs/config_1.34G_dp4_tp32_pp1_acc32_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp32_pp1_acc32_mbs4_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp4_pp1_acc32_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp4_pp1_acc32_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..06829453fb3b920014d349cd5294aab329fc8528 --- /dev/null +++ b/configs/config_1.34G_dp4_tp4_pp1_acc32_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp4_pp1_acc32_mbs16_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp64_pp1_acc16_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp64_pp1_acc16_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bc740903a61942ba32727ac39fe191dbb1b3dd02 --- /dev/null +++ b/configs/config_1.34G_dp4_tp64_pp1_acc16_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp64_pp1_acc16_mbs2_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp64_pp1_acc8_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp64_pp1_acc8_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b858001d21d6ef96a6915c7f61d448b962ec8d11 --- /dev/null +++ b/configs/config_1.34G_dp4_tp64_pp1_acc8_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp64_pp1_acc8_mbs16_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp64_tp1_pp2_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp64_tp1_pp2_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bc109e126b234c00917eff039369fdf144ddd91c --- /dev/null +++ b/configs/config_1.34G_dp64_tp1_pp2_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp64_tp1_pp2_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp2_pp16_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp2_pp16_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6da56e091c0b0360e022dc248188ed1516fa13fc --- /dev/null +++ b/configs/config_1.34G_dp8_tp2_pp16_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp2_pp16_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp4_pp1_acc1_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp4_pp1_acc1_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5d3f8090004437ec871b07aded30f82640ae3a61 --- /dev/null +++ b/configs/config_1.34G_dp8_tp4_pp1_acc1_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp4_pp1_acc1_mbs16_seq32768_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp8_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp8_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..53aba41b1441136375e0aa438867c788fff98944 --- /dev/null +++ b/configs/config_1.34G_dp8_tp8_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp8_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp16_pp1_acc1_mbs40_seq4096_zero0_tpmodeALL_vocab131k_cache.yaml b/configs/config_3.57G_dp1_tp16_pp1_acc1_mbs40_seq4096_zero0_tpmodeALL_vocab131k_cache.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c277b1b1649b7a26a1dbf9b66bf09f4665908a0e --- /dev/null +++ b/configs/config_3.57G_dp1_tp16_pp1_acc1_mbs40_seq4096_zero0_tpmodeALL_vocab131k_cache.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp16_pp1_acc1_mbs40_seq4096_zero0_tpmodeALL_vocab131k_cache + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 40 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp32_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp1_tp32_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cdcf27dd6a687817826b3912ec1cd622b932e4e1 --- /dev/null +++ b/configs/config_3.57G_dp1_tp32_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp32_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp4_pp8_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp1_tp4_pp8_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a75ae11f4cb1462167ec1014829808959e1eff5 --- /dev/null +++ b/configs/config_3.57G_dp1_tp4_pp8_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp4_pp8_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp8_pp1_acc1_mbs26_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml b/configs/config_3.57G_dp1_tp8_pp1_acc1_mbs26_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml new file mode 100644 index 0000000000000000000000000000000000000000..00c140f87632c188a4cc1258b9e8ea62adca4dfa --- /dev/null +++ b/configs/config_3.57G_dp1_tp8_pp1_acc1_mbs26_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp8_pp1_acc1_mbs26_seq4096_zero0_tpmodeRED_vocab131k_cache + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 26 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp8_pp1_acc1_mbs2_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml b/configs/config_3.57G_dp1_tp8_pp1_acc1_mbs2_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bac1919ad4c9d92580f29ed62432cf93d41fd55b --- /dev/null +++ b/configs/config_3.57G_dp1_tp8_pp1_acc1_mbs2_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp8_pp1_acc1_mbs2_seq4096_zero0_tpmodeRED_vocab131k_cache + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp1_pp16_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp1_pp16_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fc9db33dd850fdbc8cbc66e9ec24418af7069cae --- /dev/null +++ b/configs/config_3.57G_dp2_tp1_pp16_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp1_pp16_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp1_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp1_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eefe1b2586983e7a88e71ec5cecdfb830b229a2f --- /dev/null +++ b/configs/config_3.57G_dp2_tp1_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp1_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp4_pp2_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp4_pp2_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d3f7d4e3f5a4736fcd4af045a03d3ab18c622c72 --- /dev/null +++ b/configs/config_3.57G_dp2_tp4_pp2_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp4_pp2_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp4_tp16_pp4_acc4_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp4_tp16_pp4_acc4_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4494856ec20ed00c56d7151ffaddf22cc84262ac --- /dev/null +++ b/configs/config_3.57G_dp4_tp16_pp4_acc4_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp4_tp16_pp4_acc4_mbs16_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp4_tp32_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp4_tp32_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a866f20a641be2e0e07bd74c7b577e41db8da662 --- /dev/null +++ b/configs/config_3.57G_dp4_tp32_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp4_tp32_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp64_tp4_pp1_acc2_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp64_tp4_pp1_acc2_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..92c0ab7ae092b77fbeed38361129ede9a7d6e683 --- /dev/null +++ b/configs/config_3.57G_dp64_tp4_pp1_acc2_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp64_tp4_pp1_acc2_mbs32_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp16_tp16_pp1_acc16_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp16_tp16_pp1_acc16_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ad7e9838bbc162c04528a11506d6a130bcf68bd8 --- /dev/null +++ b/configs/config_469G_dp16_tp16_pp1_acc16_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp16_tp16_pp1_acc16_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp2_tp8_pp2_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp2_tp8_pp2_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..92a50aba1bdf7d29562089f8b51dfc45b504caa4 --- /dev/null +++ b/configs/config_469G_dp2_tp8_pp2_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp2_tp8_pp2_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp4_tp16_pp2_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp4_tp16_pp2_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9551ccd73aa6c574e518558016d69820035675c4 --- /dev/null +++ b/configs/config_469G_dp4_tp16_pp2_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp4_tp16_pp2_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp4_tp1_pp2_acc2_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp4_tp1_pp2_acc2_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..19ffa2aadf72eb46f90d9d92f754f81848204d67 --- /dev/null +++ b/configs/config_469G_dp4_tp1_pp2_acc2_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp4_tp1_pp2_acc2_mbs32_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp4_tp1_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp4_tp1_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..930afeb528deefacd5c02de6c7e77bda8264ddb4 --- /dev/null +++ b/configs/config_469G_dp4_tp1_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp4_tp1_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp16_tp16_pp1_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp16_tp16_pp1_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7f268b458753547bbbd628abf0fa7f16146bae71 --- /dev/null +++ b/configs/config_8.86G_dp16_tp16_pp1_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp16_tp16_pp1_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp16_pp16_acc64_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp16_pp16_acc64_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b3928d8d06fb6879872928c5176753591cac31e1 --- /dev/null +++ b/configs/config_8.86G_dp1_tp16_pp16_acc64_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp16_pp16_acc64_mbs4_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp16_pp8_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp16_pp8_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b416546aa781781072726c10fdecbfa4bda2a8f0 --- /dev/null +++ b/configs/config_8.86G_dp1_tp16_pp8_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp16_pp8_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 256 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp2_tp16_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp2_tp16_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..759812f0d1eeb19fe444b74f898a33986d7dacd8 --- /dev/null +++ b/configs/config_8.86G_dp2_tp16_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp2_tp16_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp32_tp4_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp32_tp4_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4f669affbd5f394c4265fa5d57eae821b300e260 --- /dev/null +++ b/configs/config_8.86G_dp32_tp4_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp32_tp4_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp32_tp4_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp32_tp4_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9285e1bd40f3f19352986e23c774b3921e96fba9 --- /dev/null +++ b/configs/config_8.86G_dp32_tp4_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp32_tp4_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp4_tp32_pp2_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp4_tp32_pp2_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2337120bae427e4ff1f6ce97e7657b3a3fe5d5e4 --- /dev/null +++ b/configs/config_8.86G_dp4_tp32_pp2_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp4_tp32_pp2_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp4_tp4_pp8_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp4_tp4_pp8_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6444c244c24605b408bf728aa4f1ce95178e1c8b --- /dev/null +++ b/configs/config_8.86G_dp4_tp4_pp8_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp4_tp4_pp8_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp4_tp8_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp4_tp8_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b5c8fd81815f15969a602bfbc74542beab1f08ee --- /dev/null +++ b/configs/config_8.86G_dp4_tp8_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp4_tp8_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp8_tp1_pp16_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp8_tp1_pp16_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..30a6214388323d117278b13f34b1370447c77efa --- /dev/null +++ b/configs/config_8.86G_dp8_tp1_pp16_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp8_tp1_pp16_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp16_tp4_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp16_tp4_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..013bdf7dcf05918f1de1966f51a436b34b6c5b4a --- /dev/null +++ b/configs/config_80G_dp16_tp4_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp16_tp4_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp16_tp8_pp2_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp16_tp8_pp2_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d29f72d7a2b8b80c152a7aee0b8dae43a386c724 --- /dev/null +++ b/configs/config_80G_dp16_tp8_pp2_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp16_tp8_pp2_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp1_tp32_pp2_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp32_pp2_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bffa26f731c42501a662fa454724d875f79df8a8 --- /dev/null +++ b/configs/config_80G_dp1_tp32_pp2_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp1_tp32_pp2_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 256 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp1_tp8_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp8_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..aa6e9fe36c018a5ec6dc98ea6d17b0468fee46aa --- /dev/null +++ b/configs/config_80G_dp1_tp8_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp1_tp8_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp8_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp8_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7790b04b6dd575bc3470b4cfc34e44106687e844 --- /dev/null +++ b/configs/config_80G_dp2_tp8_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp8_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 32 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp64_tp4_pp2_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp64_tp4_pp2_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..28a9b0de889625977794461101559b46f5a49b0a --- /dev/null +++ b/configs/config_80G_dp64_tp4_pp2_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp64_tp4_pp2_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp8_tp4_pp2_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp8_tp4_pp2_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ae710863e4eba71522818d8f2e22a9305e955108 --- /dev/null +++ b/configs/config_80G_dp8_tp4_pp2_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp8_tp4_pp2_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_dp1_tp8_pp1_acc1_mbs1_seq2048_zero0.yaml b/configs/config_dp1_tp8_pp1_acc1_mbs1_seq2048_zero0.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c592495edd8a1c994c5eee68a14d532c9d97fbab --- /dev/null +++ b/configs/config_dp1_tp8_pp1_acc1_mbs1_seq2048_zero0.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: dp1_tp8_pp1_acc1_mbs1_seq2048_zero0 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 24 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 49152 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_dp2_tp8_pp1_acc1_mbs8_seq2048_zero0.yaml b/configs/config_dp2_tp8_pp1_acc1_mbs8_seq2048_zero0.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0b57263ee3871bfa365faebf39b47f0d14789c0f --- /dev/null +++ b/configs/config_dp2_tp8_pp1_acc1_mbs8_seq2048_zero0.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: dp2_tp8_pp1_acc1_mbs8_seq2048_zero0 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 24 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 49152 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_dp8_tp1_pp1_acc1_mbs1_seq4096_zero0_l24_h2048_heads16.yaml b/configs/config_dp8_tp1_pp1_acc1_mbs1_seq4096_zero0_l24_h2048_heads16.yaml new file mode 100644 index 0000000000000000000000000000000000000000..00cdc586e0f6841934727126e44c29f01e696990 --- /dev/null +++ b/configs/config_dp8_tp1_pp1_acc1_mbs1_seq4096_zero0_l24_h2048_heads16.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: dp8_tp1_pp1_acc1_mbs1_seq4096_zero0_l24_h2048_heads16 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 16 + num_hidden_layers: 24 + num_key_value_heads: 16 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100