xtuner

/

llava-llama-3-8b-v1_1-transformers

like 40

i was testing it parallel to llava-hf/llava-1.5-7b-hf, and noticed that the img-to-text wasn't performing as well as the llava-hf/llava-1.5-7b-hf. this model seems to be taking much longer to generate the response and the results aren't as stellar... i have tried modifying the max_tokens value, didn't help. not sure why it scored higher in the benchmarks compared to the llava-hf/llava-1.5-7b-hf.