Farellis's picture
fine-tuned SigLIP vision tower (torch) for GPU re-embed
f9b1372 verified
Raw
History Blame Contribute Delete
429 Bytes
{
"architectures": [
"SiglipVisionModel"
],
"attention_dropout": 0.0,
"dtype": "float32",
"hidden_act": "gelu_pytorch_tanh",
"hidden_size": 768,
"image_size": 512,
"intermediate_size": 3072,
"layer_norm_eps": 1e-06,
"model_type": "siglip_vision_model",
"num_attention_heads": 12,
"num_channels": 3,
"num_hidden_layers": 12,
"patch_size": 16,
"transformers_version": "4.57.6"
}