fokan/medsiglip-448-int8
INT8 dynamic quantized version of google/medsiglip-448
- Quantization: dynamic INT8 on all nn.Linear layers (PyTorch)
- Intended for CPU inference & smaller disk footprint
- Saved as
pytorch_model.bin(quantized weights); config & processor included.
Note: Quantized state_dict is stored with PyTorch serialization (not safetensors) due to quantization tensors.