feat(server): Using quantize_config.json instead of GPTQ_BITS env variables.
#671
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Narsil
commented
Jul 21, 2023
| num_heads = config.num_attention_heads | ||
|
|
||
| self.embed_dim = embed_dim | ||
| self.hidden_size = hidden_size |
There was a problem hiding this comment.
Not really related. but this is a bug in current OPT code that I found out. I can roll this back if you want.
4 tasks
variables.
- Current PR is not great because we're side stepping the
`Weights.__init__` but Weights shouldn't requires anything related
to the config or the model_id as it aims to be a simple Wrapper
over multi file loading.
- Ideal solution would be to use something like Rust enum
```
enum Quantize{
Bitandbytes(Bitsandbytes),
GPTQ(bits: usize, groupsize: usize)
```
And passing that around during load. Unfortunately we don't
have access to this, so for now, side-stepping seems easier.
Narsil
force-pushed
the
using_quantize_config
branch
from
1fb5f88
to
0635d0e
Compare
|
Merging. It's only the flaky tests from exllama kernels. We're fixing their flakyness in a different PR. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Current PR is not great because we're side stepping the
Weights.__init__but Weights shouldn't requires anything relatedto the config or the model_id as it aims to be a simple Wrapper
over multi file loading.
Ideal solution would be to use something like Rust enum
And passing that around during load. Unfortunately we don't
have access to this, so for now, side-stepping seems easier.
Re-enabling groupsize<0 with exllama (confirmed it works.)
Helps #601
In next steps we should make sure our quantization script uses that format and make it standard.
What does this PR do?
Fixes # (issue)
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.