[Bugfix] Fix dummy weight for fp8#4916
Conversation
comaniac
left a comment
comaniac
left a comment
There was a problem hiding this comment.
LGTM. Please fix the CI. Looks like the test model outputs were changed.
| for param in model.state_dict().values(): | ||
| if torch.is_floating_point(param): | ||
| param.data.uniform_(low, high) | ||
| if torch.finfo(param.data.dtype).bits < 16: |
There was a problem hiding this comment.
Can you add a comment saying we do this because uniform doesn't support <8-bit data type?
There was a problem hiding this comment.
Sure, Added a comment :) Thanks
|
cc @robertgshaw2-neuralmagic @mgoin |
| if torch.finfo(param.data.dtype).bits < 16: | ||
| # uniform_ doesn't support < 16-bit datatypes (FP8) | ||
| dtype = param.data.dtype | ||
| param.data.to(torch.float16).uniform_(low, high).to(dtype) |
There was a problem hiding this comment.
I don't think this is doing what you want it to do -- it creates an FP16 copy, which is then modified. But the original tensor is kept unchanged:
import torch
z = torch.zeros(3)
z.to(torch.float16).uniform_(-0.1, 1.0)
In [31]: z
Out[31]: tensor([0., 0., 0.])In order to have this take effect, I believe you would need to assign it param.data at the end I believe :)
There was a problem hiding this comment.
You're right, I've missed that. fixed the PR, now it assigns the copy's data back toparam.data .
data will be assigned back to param.data
pcmoritz
left a comment
pcmoritz
left a comment
There was a problem hiding this comment.
LGTM! I'll manually test it before merging :)
|
Thanks for the contribution, it seems to be working for me 🎉 |
Allow dummy load format for fp8, torch.uniform_ doesn't support FP8 at the moment Co-authored-by: Mor Zusman <morz@ai21.com>
Allow dummy load format for fp8, torch.uniform_ doesn't support FP8 at the moment Co-authored-by: Mor Zusman <morz@ai21.com>
Allow dummy load format for fp8, torch.uniform_ doesn't support FP8 at the moment Co-authored-by: Mor Zusman <morz@ai21.com>
3 participants
Allow dummy load format for fp8,
torch.uniform_doesn't support FP8 at the momentBEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE
PR Checklist (Click to Expand)
Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.
PR Title and Classification
Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:
[Bugfix]for bug fixes.[CI/Build]for build or continuous integration improvements.[Doc]for documentation fixes and improvements.[Model]for adding a new model or improving an existing model. Model name should appear in the title.[Frontend]For changes on the vLLM frontend (e.g., OpenAI API server,LLMclass, etc.)[Kernel]for changes affecting CUDA kernels or other compute kernels.[Core]for changes in the core vLLM logic (e.g.,LLMEngine,AsyncLLMEngine,Scheduler, etc.)[Hardware][Vendor]for hardware-specific changes. Vendor name should appear in the prefix (e.g.,[Hardware][AMD]).[Misc]for PRs that do not fit the above categories. Please use this sparingly.Note: If the PR spans more than one category, please include all relevant prefixes.
Code Quality
The PR need to meet the following code quality standards:
format.shto format your code.docs/source/if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.Notes for Large Changes
Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with
rfc-requiredand might not go through the PR.What to Expect for the Reviews
The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:
action-requiredlabel on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.Thank You
Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!