xiaolesu/Lean4-sft-nt-8b

See axolotl config

axolotl version: 0.15.0.dev0

base_model: Qwen/Qwen3-8B

load_in_8bit: false
load_in_4bit: false
strict: false

plugins:
  - axolotl.integrations.liger.LigerPlugin

liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_layer_norm: true
liger_fused_linear_cross_entropy: true

chat_template: qwen3
chat_template_kwargs:
  enable_thinking: false

datasets:
  - path: xiaolesu/lean4-sft-stmt
    type: alpaca
    split: train
  - path: xiaolesu/lean4-sft-stmt
    type: alpaca
    split: validation

output_dir: ./outputs/qwen3-sft-stmt/

sequence_len: 4096
sample_packing: true
flex_attention: true

flex_attn_compile_kwargs:
  dynamic: false
  mode: max-autotune-no-cudagraphs

wandb_project: qwen3-sft-stmt
wandb_entity:
wandb_watch:
wandb_name: qwen3-8b-run1
wandb_log_model:

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

bf16: true
tf32: true

resume_from_checkpoint:
logging_steps: 5

evals_per_epoch: 10
saves_per_epoch: 10
save_total_limit: 3

warmup_ratio: 0.1
weight_decay: 0.0
fsdp:
  - full_shard
  - auto_wrap

fsdp_config:
  fsdp_version: 2
  fsdp_offload_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
  fsdp_reshard_after_forward: true
  fsdp_activation_checkpointing: true

special_tokens:

outputs/qwen3-sft-stmt/

This model is a fine-tuned version of Qwen/Qwen3-8B on the xiaolesu/lean4-sft-stmt and the xiaolesu/lean4-sft-stmt datasets.

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: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 16
• total_eval_batch_size: 16
• 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: 16
• training_steps: 168

Training results

Framework versions

• Transformers 5.2.0
• Pytorch 2.9.1+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2