qnguyen3/Quan-Mathy-7B

See axolotl config

axolotl version: 0.5.0

base_model: Qwen/Qwen2.5-Math-7B

plugins:
  - axolotl.integrations.liger.LigerPlugin

liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: true

strict: false

chat_template: chatml
datasets:

  - path: arcee-ai/orcamath_evol_85k
    type: chat_template
    split: train
    field_messages: conversations
    message_field_role: from
    message_field_content: value

  - path: allenai/tulu-3-sft-personas-math
    type: chat_template
    split: train[:10%]
    field_messages: messages
    message_field_role: role
    message_field_content: content

  - path: allenai/tulu-3-sft-personas-algebra
    type: chat_template
    split: train
    field_messages: messages
    message_field_role: role
    message_field_content: content

dataset_prepared_path: ./axolotl-datasets/math-evol-prepared
val_set_size: 0.02
output_dir: ./axolotl-outputs/Arcee-7B-Mathy-7B-6e

sequence_len: 4096
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

wandb_project: "Arcee-Mathy-7B"
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 4
num_epochs: 6
optimizer: adamw_torch_fused #adamw_torch_fused # if you have OOM errors you can use adamw_8bit
lr_scheduler: linear
learning_rate: 5e-6

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 50
evals_per_epoch: 1
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
weight_decay: 0.0

special_tokens:
  pad_token: <|endoftext|>
  eos_token: <|im_end|>

axolotl-outputs/Arcee-7B-Mathy-7B-6e

This model is a fine-tuned version of Qwen/Qwen2.5-Math-7B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.5608

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-06
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 256
• total_eval_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 6

Training results

Training Loss	Epoch	Step	Validation Loss
0.4101	0.0106	1	1.6490
0.2319	0.9987	94	1.5007
0.2234	1.9960	188	1.5070
0.205	2.9920	282	1.5350
0.1979	3.9894	376	1.5456
0.1866	4.9867	470	1.5547
0.1926	5.9827	564	1.5608

Framework versions

• Transformers 4.46.1
• Pytorch 2.3.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.3