iocuydi

/

amharic-llava

like 3

Unfortunately I think it'll be a bit harder to use hf inference. This model has a few dependencies to work properly:
-llama2 weights
-pretrained amharic llava weights
-clip weights
-must load the amharic llava weights but ensure the naming/overwriting is correct as done here (https://github.com/iocuydi/amharic-llama-llava/blob/c8e86f065d0f5411223b492c8ae0b0b8cad4cffc/llava/llava/model/builder.py#L25)

Haven't used hf for hosted inference before but feel free to update it if this is easily workable.
You can also run it locally by following the steps here https://github.com/iocuydi/amharic-llama-llava/tree/main/llava