-
Notifications
You must be signed in to change notification settings - Fork 314
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
unknown dtype for tensor (BF16?) #663
Comments
@oldgithubman yes, this is the problem. Please see huggingface/candle#2387. This will enable support for BF16 and more descriptive errors! |
@oldgithubman given that the Candle PR hasn't been merged, I have mirrored my changes onto our Candle fork so we can proceed. Please see #691, which should enable this to work. To test:
|
|
@oldgithubman thanks, that should be fixed now if you |
|
@oldgithubman can you please run with |
that was run with RUST_BACKTRACE=full. Do you still want me to do it with 1? |
Ah ok thanks, I'll take a look. |
@oldgithubman I just updated the branch to correctly setup the QMatMul (#691). |
Works! |
@oldgithubman thanks for confirming! I just merged #691, so this feature is available on |
@oldgithubman closing this issue as it works, please feel free to reopen! |
Describe the bug
My Q8_0 quant of Athene-70B loads fine. I have another quant that is identical except the output and embedding tensors are BF16:
Latest commit or version
0.2.4
The text was updated successfully, but these errors were encountered: