| { | |
| "project": "LLM-Based Multimodal Behaviour-Change Support", | |
| "training_script": "models/audio/train.py", | |
| "training_output_dir": "C:\\hackathon\\Gemini_CLI\\EMORA\\models\\audio\\runs_emotion2vec", | |
| "backbone_hub_id": "iic/emotion2vec_base_finetuned", | |
| "backbone_loader": "funasr.AutoModel", | |
| "head_reconstruction": "At inference time the container rebuilds the nn.Sequential head from src/pipeline/audio_pipeline.py and loads this state dict into it. The backbone is fetched from Hugging Face Hub at runtime.", | |
| "label_mapping_note": "Audio was trained directly on the 6-class shared space: angry, disgust, fearful, happy, sad, surprised.", | |
| "files": { | |
| "model_state_dict.pt": "trained classification head state dict", | |
| "config.json": "reconstruction metadata for the head" | |
| } | |
| } | |