| run_id: xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu_lora_fsdp_full_train_noval_tail_logits_fullstatesave_v6 | |
| stage: qwen_lora_text_video_audio | |
| backbone_id: qwen3_omni_lora | |
| dataset_contract: xperience10m_episode_json_qa_v1 | |
| model_id: <model-cache>/Qwen__Qwen3-Omni-30B-A3B-Instruct | |
| dataset_jsonl: results/omni_finetune/xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu_dataset/dataset.jsonl | |
| checkpoint_dir: <project>/checkpoints/xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu_lora_fsdp_full_train_noval_tail_logits_fullstatesave_v6/adapter_lora | |
| num_processes: 8 | |
| epochs: 1 | |
| learning_rate: 0.0001 | |
| lora_r: 16 | |
| lora_alpha: 32 | |
| loss_mode: answer_token_ce | |
| loss_logit_tail_only: True | |