microsoft

/

Phi-3.5-vision-instruct

like 463

Unfortunately, llama.cpp currently does not support some techs like LongRoPE, Flash attention, CLIP ViT projector, etc which used by Phi-3... Ollama uses llama.cpp as infra. So...

Unfortunately, llama.cpp currently does not support some techs like LongRoPE, Flash attention, CLIP ViT projector, etc which used by Phi-3... Ollama uses llama.cpp as infra. So...

Looks like I made a mistake... llava-phi3 on ollama can use CLIP ViT model as part of gguf model files. And combine with the Phi-3 text model. I just meet issues when I try to convert Phi-3/3.5 vision to GGUF by llama.cpp.