mixvideo-v2/cargos/tvai-v2/视觉-语言模型配置/ViT-L-16-SigLIP2-512.json

32 lines
788 B
JSON

{
"embed_dim": 1024,
"init_logit_bias": -10,
"custom_text": true,
"vision_cfg": {
"image_size": 512,
"timm_model_name": "vit_large_patch16_siglip_512",
"timm_model_pretrained": false,
"timm_pool": "map",
"timm_proj": "none"
},
"text_cfg": {
"context_length": 64,
"vocab_size": 256000,
"hf_tokenizer_name": "timm/ViT-L-16-SigLIP2-512",
"tokenizer_kwargs": {
"clean": "canonicalize"
},
"width": 1024,
"heads": 16,
"layers": 24,
"no_causal_mask": true,
"proj_bias": true,
"pool_type": "last",
"norm_kwargs":{
"eps": 1e-6
},
"act_kwargs": {
"approximate": "tanh"
}
}
}