Chinese-LLaMA-Alpaca VS llm-leaderboard

I'd like to share with you today the Chinese-Alpaca-Plus-13B-GPTQ model, which is the GPTQ format quantised 4bit models of Yiming Cui's Chinese-LLaMA-Alpaca 13B for GPU reference.

Could follow what people did with the Chinese-LLaMA, just for Korean. Might want to have a pure Korean corpus before feeding in a translation dataset. How big is it by the way?

Chinese: https://github.com/ymcui/Chinese-LLaMA-Alpaca

This is assuming you’re using a really big LLM behind a paid service. There are plenty of smaller open source models. Not sure at what point it’s not “large” but when fine tuned they are capable of matching the largest LLM in performance on narrow tasks.

Some of these open source models can even be run on your local machine. It’d be very inexpensive to run thousands of pages through it.

https://llm-leaderboard.streamlit.app/

palm-2-l-instruct scores 0.909 on Winogrande few-shot.

https://github.com/LudwigStumpp/llm-leaderboard/blob/main/RE...

They said it "rivals OpenAI’s Codex model" which performs worse than starcoder-16b on HumanEval-Python (pass@1) according to https://github.com/LudwigStumpp/llm-leaderboard

Yes, see this leaderboard for a comparison: https://llm-leaderboard.streamlit.app/

So I've tried to use a basic matrix factorization method to estimate unknown benchmark scores for models based on the known benchmark scores. Basically, I assume each model has some intrinsic "quality" score, and the known benchmarks are assumed to be a linear function of the quality score. This is similar to matrix factorization with only 1 latent factor (though the bias values have to handled differently). Then I fit the known benchmark scores from https://github.com/LudwigStumpp/llm-leaderboard to my parameters, and estimate the remaining benchmark scores.