{ "architectures": [ "AVSDForResponseGeneration" ], "audio_dim": 128, "decoder_layers": 1, "decoder_proj": 1024, "dropout": 0.5, "dtype": "float32", "embed_size": 128, "eos_token_id": 1, "frame_dim": 512, "hidden_size": 256, "history_dim": 128, "history_hidden": 128, "history_pair_layers": 1, "history_word_layers": 2, "max_answer_length": 20, "max_question_length": 10, "model_type": "fga_avsd_generation", "num_audio_steps": 32, "num_frame_regions": 49, "num_video_frames": 0, "num_video_regions": 49, "num_video_streams": 4, "pad_token_id": 0, "question_dim": 256, "question_layers": 1, "tie_embeddings": false, "transformers_version": "5.14.1", "use_cache": false, "use_sizes": false, "video_dim": 512, "vocab_size": 5645 }