AlpacaEval results by chatgpt! Community run
Hi,
Since phi2 requires remote code which HF open llm leaderboard would not accept at this moment,
I ran phi2 and my sft https://huggingface.co/Yhyu13/phi-2-sft-alpaca_gpt4_en-ep1-lora/tree/main on alpaca gpt4 en to the AlpcaEval benchmark
https://tatsu-lab.github.io/alpaca_eval/
Here is result evaluated by chatpgpt https://github.com/tatsu-lab/alpaca_eval/pull/183
win_rate standard_error n_total avg_length
gpt4 73.79 1.54 805 1365
claude 70.37 1.60 805 1082
chatgpt 66.09 1.66 805 811
wizardlm-13b 65.16 1.67 805 985
vicuna-13b 64.10 1.69 805 1037
guanaco-65b 62.36 1.71 805 1249
oasst-rlhf-llama-33b 62.05 1.71 805 1079
alpaca-farm-ppo-human 60.25 1.72 805 803
falcon-40b-instruct 56.52 1.74 805 662
phi-2-alpaca-gpt4(new) 54.23 1.75 804 1138
text_davinci_003 50.00 0.00 805 307
alpaca-7b 45.22 1.74 805 396
phi-2(new) 43.79 1.74 805 924
text_davinci_001 28.07 1.56 805 296
It could be a milestone for small models, we finally have one open model can run for everyone which surpass GPT3.5!
Cheers!
Wow, nice to see a community replication. I place a lot of weight on these, as they are independent. Looks like it's conformed to be best-for-size.
It will be interesting if someone uses Claude as a judge. Since GPT4 was presumably used to generate the training data, so there is a little bit of a bias when used as judge (of course it likes its own student, it says exactly what it would say!).