Fine-tuned model (merged)

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Dataset: u-10bei/sft_alfworld_trajectory_dataset_v5
• Method: SFT (assistant-only loss)
• Format: Merged full model

Training config (key)

• max_seq_len: 1024
• epochs: 1
• per_device_train_bs: 1
• grad_accum: 4
• lr: 2e-06
• warmup_ratio: 0.1
• weight_decay: 0.05
• lora_r: 64
• lora_alpha: 128
• lora_dropout: 0.0
• target_modules: q_proj,k_proj,v_proj,o_proj,gate_proj,up_proj,down_proj

Generated by train_sft.py.