fblgit/TheBeagle-v2beta-32B-MGS failing evaluation
Hi @clefourrier
Not exactly sure the nature of the failure evaluating fblgit/TheBeagle-v2beta-32B-MGS both bf16 and fp16.
I ran lm-eval tests with no issues on my lab, followed also the config/tokenizer/model load-up test also with no issue.
I also ran the lm-eval evaluations of the leaderboard with no issues as well.
Can you please share with me whats the error that its causing FAILED status on the evaluation over there?
PD: If the code base is opensauce, I can push a PR to incorporate some verbosity on the COMMIT msg when a failure is found, in a way that doesn't mess the JSON payload structure of the eval pipeline.
Thx in advance, best regards.
Hi @fblgit ,
Please, do not open discussion in Requests dataset as we don't follow them and may miss them. It will be way better to create a discussion here in the Community section next time
Anyway, I would be happy to help you with your issue now! According to our FAQ, could you please provide requests files for your model both in bf16 and f16?
https://huggingface.co/datasets/open-llm-leaderboard/requests/blob/main/fblgit/TheBeagle-v2beta-32B-MGS_eval_request_False_bfloat16_Original.json
https://huggingface.co/datasets/open-llm-leaderboard/requests/blob/main/fblgit/TheBeagle-v2beta-32B-MGS_eval_request_False_float16_Original.json
I'll follow ur advise next time and open it in the right section, appreciate your help.
Best Regards
Thank you for providing request files! I've checked your models, the problem was on our side and it's fixed now, you can find your models on the Leaderboard
I'm closing this issue, please, feel free to open a new one!