You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi @francistotle , thanks for pointing that out!
I accidentally uploaded the wrong files earlier, it was indeed the 7B model. I've now updated the repository with the correct 32B version. Let me know if you run into any further issues!
We used a context length of 131072 (i.e., 128k) for both the 7B and 32B models.
For better results with the 32B model, make sure to set max_seq_length=131072 during inference.
Activity
dc4818773 commentedon Jun 25, 2025
francistotle commentedon Jun 25, 2025
the 32 b you released here https://huggingface.co/czlll/Qwen2.5-Coder-32B-CL is a 7b model. PResumably the same one as the 7b release
francistotle commentedon Jun 27, 2025
Let me know if you are able to update this!
czlll commentedon Jun 28, 2025
Hi @francistotle , thanks for pointing that out!
I accidentally uploaded the wrong files earlier, it was indeed the 7B model. I've now updated the repository with the correct 32B version. Let me know if you run into any further issues!
francistotle commentedon Jul 2, 2025
Thanks, what context length did you use for your results btw?
francistotle commentedon Jul 3, 2025
I think the 32b model doesn't work very well, are you sure you uploaded the right checkpoint? I get much higer performance with the 7b version
czlll commentedon Jul 16, 2025
We used a context length of 131072 (i.e., 128k) for both the 7B and 32B models.
For better results with the 32B model, make sure to set max_seq_length=131072 during inference.