{
  "base_model": "Qwen/Qwen3.5-9B",
  "audio_encoder": "openai/whisper-medium",
  "dataset": "Joysw909/AVQA",
  "max_train_samples": 2048,
  "max_eval_samples": 256,
  "max_text_len": 384,
  "max_audio_seconds": 12,
  "num_audio_tokens": 8,
  "audio_projector_hidden_dim": 512,
  "audio_projector_depth": 1,
  "batch_size": 1,
  "grad_accum": 16,
  "learning_rate": 0.0001,
  "epochs": 1,
  "low_vram_safe_mode": true,
  "qwen_backbone_mode": "full_bnb_lora",
  "push_to_hub": true,
  "hf_repo_id": "Hellohihihih/qwen35-audio-graft-avqa"
}