Patricijia/smolvlm-gif-descriptor

v7: base vision/embed + patched decoder only (safe)

v6: fixed transpose detection for square matrices in decoder

v5: name-based decoder patching (verified different outputs in fp16)

All ONNX files: base model graph + value-matched LoRA weight patches (156 tensors)

ONNX with merged LoRA: patched vision encoder + exported decoder

Delete onnx/vision_encoder_q4.onnx with huggingface_hub

Delete onnx/vision_encoder_int8.onnx with huggingface_hub

Delete onnx/vision_encoder_fp16.onnx with huggingface_hub

Delete onnx/vision_encoder_bnb4.onnx with huggingface_hub

Delete onnx/vision_encoder.onnx with huggingface_hub

Delete onnx/embed_tokens_uint8.onnx with huggingface_hub

Delete onnx/embed_tokens_quantized.onnx with huggingface_hub

Delete onnx/embed_tokens_q4f16.onnx with huggingface_hub

Delete onnx/embed_tokens_q4.onnx with huggingface_hub

Delete onnx/embed_tokens_int8.onnx with huggingface_hub

Delete onnx/embed_tokens_fp16.onnx with huggingface_hub

Delete onnx/embed_tokens_bnb4.onnx with huggingface_hub

Delete onnx/embed_tokens.onnx with huggingface_hub

Delete onnx/decoder_model_merged_uint8.onnx with huggingface_hub

Delete onnx/decoder_model_merged_quantized.onnx with huggingface_hub

Delete onnx/decoder_model_merged_q4f16.onnx with huggingface_hub

Delete onnx/decoder_model_merged_q4.onnx with huggingface_hub

Delete onnx/decoder_model_merged_int8.onnx with huggingface_hub

Delete onnx/decoder_model_merged_fp16.onnx with huggingface_hub

Delete onnx/decoder_model_merged_bnb4.onnx with huggingface_hub

Delete onnx/decoder_model_merged.onnx with huggingface_hub

Copy onnx/vision_encoder_uint8.onnx from base model (working ONNX)

Copy onnx/vision_encoder_quantized.onnx from base model (working ONNX)

Copy onnx/vision_encoder_q4f16.onnx from base model (working ONNX)

Copy onnx/vision_encoder_q4.onnx from base model (working ONNX)

Copy onnx/vision_encoder_int8.onnx from base model (working ONNX)

Copy onnx/vision_encoder_fp16.onnx from base model (working ONNX)

Copy onnx/vision_encoder_bnb4.onnx from base model (working ONNX)

Copy onnx/vision_encoder.onnx from base model (working ONNX)

Copy onnx/embed_tokens_uint8.onnx from base model (working ONNX)

Copy onnx/embed_tokens_quantized.onnx from base model (working ONNX)

Copy onnx/embed_tokens_q4f16.onnx from base model (working ONNX)

Copy onnx/embed_tokens_q4.onnx from base model (working ONNX)

Copy onnx/embed_tokens_int8.onnx from base model (working ONNX)

Copy onnx/embed_tokens_fp16.onnx from base model (working ONNX)

Copy onnx/embed_tokens_bnb4.onnx from base model (working ONNX)

Copy onnx/embed_tokens.onnx from base model (working ONNX)

Copy onnx/decoder_model_merged_uint8.onnx from base model (working ONNX)

Copy onnx/decoder_model_merged_quantized.onnx from base model (working ONNX)

Copy onnx/decoder_model_merged_q4f16.onnx from base model (working ONNX)

Copy onnx/decoder_model_merged_q4.onnx from base model (working ONNX)

Copy onnx/decoder_model_merged_int8.onnx from base model (working ONNX)

Copy onnx/decoder_model_merged_fp16.onnx from base model (working ONNX)

Copy onnx/decoder_model_merged_bnb4.onnx from base model (working ONNX)

Copy onnx/decoder_model_merged.onnx from base model (working ONNX)