{ "architectures": [ "Idefics2ForConditionalGeneration" ], "dtype": "float32", "image_token_id": 99, "model_type": "idefics2", "perceiver_config": { "attention_dropout": 0.0, "hidden_act": "silu", "hidden_size": 64, "initializer_range": 0.02, "model_type": "idefics2_perceiver", "num_key_value_heads": 1, "resampler_depth": 2, "resampler_head_dim": 12, "resampler_n_heads": 2, "resampler_n_latents": 2, "rms_norm_eps": 1e-06 }, "text_config": { "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 32, "hidden_act": "silu", "hidden_size": 64, "image_token_id": 99, "initializer_range": 0.02, "intermediate_size": 56, "max_position_embeddings": 256, "model_type": "mistral", "num_attention_heads": 2, "num_hidden_layers": 2, "num_key_value_heads": 2, "pad_token_id": 0, "rms_norm_eps": 1e-06, "rope_parameters": { "rope_theta": 10000.0, "rope_type": "default" }, "sliding_window": 32, "tie_word_embeddings": false, "use_cache": true, "vocab_size": 32003 }, "tie_word_embeddings": false, "transformers_version": "5.16.0.dev0", "use_cache": false, "vision_config": { "attention_dropout": 0.1, "dropout": 0.1, "hidden_act": "gelu_pytorch_tanh", "hidden_size": 32, "image_size": 12, "initializer_range": 0.02, "intermediate_size": 32, "layer_norm_eps": 1e-06, "model_type": "idefics2_vision", "num_attention_heads": 4, "num_channels": 3, "num_hidden_layers": 2, "patch_size": 12 }, "vocab_size": 32003 }