Spaces:
Running
on
CPU Upgrade
🚩 Report: Not working
Model_id: ehristoforu/testq-32b
Model_id: ehristoforu/qwen2.5-test-32b-it
Model_id: ehristoforu/frqwen2.5-from72b-duable10layers
No custom remote code, no edits in architechture, no edits in tokenizer, only mergekit merge
All configs imported from original Qwen models, that eval success
There are already many similar problems that are not being solved:
- https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/1078
- https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/1083
- https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/1075
- https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/1062
Please finally solve this problem, as many custom models simply cannot get into the leaderboard because of this error (I would like you to also look at the error logs and solve these errors, as the problem is clearly on your side)
Thanks in advance! I hope you will pay attention to this
@alozowski , @clefourrier , any fixes?
Hi @ehristoforu ,
You can find both ehristoforu/testq-32b and ehristoforu/qwen2.5-test-32b-it on the Leaderboard now (see the screenshot below)
But there is an error for ehristoforu/frqwen2.5-from72b-duable10layers:
ValueError: Trying to set a tensor of shape torch.Size([151665, 8192]) in "weight" (which has shape torch.Size([152064, 8192])), this looks incorrect.
Seems like a mismatch in tensor shapes when loading model weights. Could you please check it on your side?
Really thank you! Have a nice day)