GusPuffy/Reasoning-Gen-8B-Lora

See axolotl config

axolotl version: 0.9.2

base_model: Qwen/Qwen3-8B

datasets:
  - path: GusPuffy/syvai-reasoning-gen-v2
    type: chat_template
    field_messages: messages
    roles_to_train: ["assistant"]

load_in_8bit: false
load_in_4bit: true
strict: false

dataset_prepared_path: last_run_prepared
val_set_size: 0.02
output_dir: ./outputs/out

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

adapter: qlora
lora_model_dir:
lora_r: 32
lora_alpha: 64
lora_dropout: 0.1
lora_target_linear: true

wandb_project: reasoning-gen
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 1
optimizer: adamw_torch_fused
lr_scheduler: cosine
learning_rate: 2e-5

bf16: auto
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_steps: 100
evals_per_epoch: 2
saves_per_epoch: 1
weight_decay: 0.0
special_tokens:

fsdp:
  - full_shard
  - auto_wrap
fsdp_config:
  fsdp_limit_all_gathers: true
  fsdp_sync_module_states: true
  fsdp_offload_params: true
  fsdp_use_orig_params: false
  fsdp_cpu_ram_efficient_loading: true
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_transformer_layer_cls_to_wrap: Qwen3DecoderLayer
  fsdp_state_dict_type: FULL_STATE_DICT
  fsdp_sharding_strategy: FULL_SHARD

outputs/out

This model is a fine-tuned version of Qwen/Qwen3-8B on the GusPuffy/syvai-reasoning-gen-v2 dataset. It achieves the following results on the evaluation set:

• Loss: 1.2201

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: 2e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• total_eval_batch_size: 4
• 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: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.8758	0.0011	1	1.5860
0.6286	0.5004	452	1.2201

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.8.0.dev20250519+cu128
• Datasets 3.5.1
• Tokenizers 0.21.1