| { | |
| "architectures": [ | |
| "RWKV7VLForConditionalGeneration" | |
| ], | |
| "text_config":{ | |
| "a_low_rank_dim": 64, | |
| "attn": null, | |
| "attn_mode": "chunk", | |
| "bos_token_id": 0, | |
| "decay_low_rank_dim": 64, | |
| "eos_token_id": 0, | |
| "fuse_cross_entropy": true, | |
| "fuse_norm": false, | |
| "gate_low_rank_dim": 128, | |
| "head_dim": 64, | |
| "hidden_act": "sqrelu", | |
| "hidden_ratio": 4.0, | |
| "hidden_size": 1024, | |
| "initializer_range": 0.006, | |
| "intermediate_size": 4096, | |
| "max_position_embeddings": 2048, | |
| "model_type": "rwkv7_vl", | |
| "norm_bias": true, | |
| "norm_eps": 1e-05, | |
| "norm_first": true, | |
| "num_heads": 32, | |
| "num_hidden_layers": 24, | |
| "tie_word_embeddings": false, | |
| "torch_dtype": "float32", | |
| "transformers_version": "4.48.0", | |
| "use_cache": true, | |
| "v_low_rank_dim": 32, | |
| "vocab_size": 65536 | |
| }, | |
| "vision_config": { | |
| "image_size": 384, | |
| "model_type": "siglip_vision_model" | |
| }, | |
| "proj_config": { | |
| "encoder_dim": 768, | |
| "project_dim" : 1024 | |
| }, | |
| "encoder_type": "siglip2" | |
| } |