-
Notifications
You must be signed in to change notification settings - Fork 977
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
Error on interactive generation #555
Comments
Thank you for the bug report. Can you check that this wasn’t inadvertently caused by #539? |
Uhm no, I'm misunderstanding something. In both main and 2189a4f , interactive seems to only work with 3-words prompts. |
Ditto for me also. |
|
Newbie alert ! - A quick test input-file & interactive run across 8 GPU's : cross_entropy.py Line 60:
For "text-gen-type": "input-file", DEBUG tensor([0, 1, 2, 3, 4, 5, 6, 7, 8], device='cuda:6') tensor([ 58, 46434, 0], device='cuda:6') DEBUG tensor([0, 1, 2, 3, 4, 5, 6, 7, 8], device='cuda:7') tensor([0, 0, 0], device='cuda:7') For "text-gen-type": "interactive", DEBUG DEBUG tensor([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
tensor([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
DEBUG DEBUG tensor([0], device='cuda:7') tensor([0], device='cuda:7') tensor([0], device='cuda:6') tensor([0], device='cuda:6') DEBUG tensor([0], device='cuda:7') DEBUG tensor([0], device='cuda:7') tensor([0], device='cuda:6') tensor([0], device='cuda:6') DEBUG DEBUG tensor([0], device='cuda:7') tensor([0], device='cuda:7') tensor([0], device='cuda:6') tensor([0], device='cuda:6') DEBUG DEBUG tensor([0], device='cuda:7') tensor([0], device='cuda:7') tensor([0], device='cuda:6') tensor([0], device='cuda:6') DEBUG tensor([0], device='cuda:7') DEBUG tensor([0], device='cuda:7') tensor([0], device='cuda:6') tensor([0], device='cuda:6') DEBUG DEBUG tensor([0], device='cuda:6') tensor([0], device='cuda:7') tensor([0], device='cuda:6') I don't understand enough about how this has been written but is it something to do with the large number of Tensor elements in the interactive mode - caused perhaps by the code not properly dimensioning the tensor array with the interactive input ? |
With that line added on my end...
The same prompt, passed in as a file:
These tensors don't look dimensioned properly to me for smaller inputs as well, looks like we might have the root cause outlined. |
See #604 for the greater prototyping effort underway to ensure that all processes have the correct context_length and context_tokens. |
This is resolved |
I'm observing behavior similar to this issue. Whenever I enter an interactive input longer than three tokens I receive an error like this.
I don't receive any errors when the length of the input does not exceed three tokens.
Environment
|
I think I'm also experiencing this. Any interactive prompt larger than 3 words has issues. |
Describe the bug
Setting
"text-gen-type": "interactive"
results in anIndexError: : shape mismatch: indexing tensors could not be broadcast together with shapes [4], [3]
. Other generation types work.To Reproduce
Steps to reproduce the behavior:
Expected behavior
Should work like non-interactive mode.
Environment (please complete the following information):
Additional context
Using ppc64le, so some libraries are not exactly as pinned. Please ignore the issue if it does not occur on more common platforms.
The text was updated successfully, but these errors were encountered: