but should work on more configurations (no need for 2 GPUs, less RAM
usage).
# What does this PR do?
Reworking the quantization script so it's still universal (not llama
specific)
but should work on more configurations (no need for 2 GPUs, less RAM
usage).
Still need to investigate the potential differences in quantization
results.
<!--
Congratulations! You've made it this far! You're not quite done yet
though.
Once merged, your PR is going to appear in the release notes with the
title you set, so make sure it's a great title that fully reflects the
extent of your awesome contribution.
Then, please replace this with a description of the change and which
issue is fixed (if applicable). Please also include relevant motivation
and context. List any dependencies (if any) that are required for this
change.
Once you're done, someone will review your PR shortly (see the section
"Who can review?" below to tag some potential reviewers). They may
suggest changes to make the code even better. If no one reviewed your PR
after a week has passed, don't hesitate to post a new comment
@-mentioning the same persons---sometimes notifications get lost.
-->
<!-- Remove if not applicable -->
Fixes # (issue)
## Before submitting
- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Did you read the [contributor
guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests),
Pull Request section?
- [ ] Was this discussed/approved via a Github issue or the
[forum](https://discuss.huggingface.co/)? Please add a link
to it if that's the case.
- [ ] Did you make sure to update the documentation with your changes?
Here are the
[documentation
guidelines](https://github.com/huggingface/transformers/tree/main/docs),
and
[here are tips on formatting
docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation).
- [ ] Did you write any new necessary tests?
## 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.
<!-- Your PR will be replied to more quickly if you can figure out the
right person to tag with @
@OlivierDehaene OR @Narsil
-->
# What does this PR do?
When passing in environment variables like gptq_bits, we still get
errors thrown from TGI because the try/catch block is catching the wrong
type of error. This PR aims to fix that.
@Narsil - let me know if this is how you want this formatted. My Python
is a little shaky, so I hope this syntax is correct.
- The code is relatively easy (just disable the checks on Embedding and
Head)
This cannot be done in the same easy fashion for hidden_dim/head_dim.
It's relatively easy on some models (classic MHA) but it would make the
other
models (MQA) much more complex, and GPTQ quantization another quite
hairy piece
of code.
# What does this PR do?
This fixes a typo and extends the GPTP_BITS environment variables
through to the second method which requires the same logic. Please let
me know if there's anything I've misunderstood in this change.
Thanks @Narsil for the original fix.
# What does this PR do?
Some models are already converted, and do not have those values in the
file, this enables users to use them with less friction.
Went for pure env based because adding flags would end up (imo) very
tedious to maintain. There's a lot of sanitation to do: those flags
would be errors if not used in conjuction with `--quantize gptq`.
Then the flags need to exist in the launcher and the server passing them
all throughout all function calls.
This PR is intended as an easy escape hatch, not the defacto method to
use gptq in TGI.
Fixes#500
This PR allows the MPT model to be loaded from local files. Without this
change, an exception will be thrown by `hf_hub_download` function if
`model_id` is a local path.
Let's start discussing implementation.
- Need to expose the quantization scripts (either included here or add
doc on how to use https://github.com/qwopqwop200/GPTQ-for-LLaMa)
- Make sure GPTQ works for multiple models (priority to Falcon).
Currently it means that every place we use `get_{tensor|sharded}` to
check for quantization.
My idea is to reintegrate as much as possible into `utils/layer.py` by
expanding `load_multi` to be a bit more generic.
This might require some thinking, but ultimately the
`qweight,qzeros,scales,g_idx` should be in a single place, and
independant of bias presence.
# What does this PR do?
<!--
Congratulations! You've made it this far! You're not quite done yet
though.
Once merged, your PR is going to appear in the release notes with the
title you set, so make sure it's a great title that fully reflects the
extent of your awesome contribution.
Then, please replace this with a description of the change and which
issue is fixed (if applicable). Please also include relevant motivation
and context. List any dependencies (if any) that are required for this
change.
Once you're done, someone will review your PR shortly (see the section
"Who can review?" below to tag some potential reviewers). They may
suggest changes to make the code even better. If no one reviewed your PR
after a week has passed, don't hesitate to post a new comment
@-mentioning the same persons---sometimes notifications get lost.
-->
<!-- Remove if not applicable -->
Fixes # (issue)
## Before submitting
- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Did you read the [contributor
guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests),
Pull Request section?
- [ ] Was this discussed/approved via a Github issue or the
[forum](https://discuss.huggingface.co/)? Please add a link
to it if that's the case.
- [ ] Did you make sure to update the documentation with your changes?
Here are the
[documentation
guidelines](https://github.com/huggingface/transformers/tree/main/docs),
and
[here are tips on formatting
docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation).
- [ ] Did you write any new necessary tests?
## 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.
<!-- Your PR will be replied to more quickly if you can figure out the
right person to tag with @
@OlivierDehaene OR @Narsil
-->
---------
Co-authored-by: Ubuntu <ubuntu@ip-172-31-41-161.ec2.internal>
Co-authored-by: OlivierDehaene <olivier@huggingface.co>
Should be more robust to shared tensors (ok when using
`from_pretrained). But forcing us to add new checks in our loading
code (since the chosen key to keep might be different from
`transformers`).
---------
Co-authored-by: Ubuntu <ubuntu@ip-172-31-41-161.ec2.internal>
# What does this PR do?
Reworked the loading logic. Idea is to use cleaner loading code:
- Remove need for `no_init_weights`
- Remove all weird `bnb_linear` and `load_weights` and
`post_load_weights`.
New code layout:
- New class `Weights` in charge of handling loading the weights from
multiple files into appropiate tensors (potentially sharded)
- TP layers now are "shells", they contain the code to know what kind of
sharding we need + eventual `all_reduce`. They do not inherit from
linear, but they contain some kind of Linear instead
- the contained linear can be either FastLinear, BnbLinear or GPTq
Linear next.
- All modeling code is explictly made for sharding, process group is
just no-ops for non sharded code (removes a lot of test cases)
---------
Co-authored-by: Ubuntu <ubuntu@ip-172-31-41-161.taildb5d.ts.net>
Co-authored-by: Ubuntu <ubuntu@ip-172-31-41-161.ec2.internal>
Co-authored-by: OlivierDehaene <olivier@huggingface.co>
Co-authored-by: OlivierDehaene <23298448+OlivierDehaene@users.noreply.github.com>
OlivierDehaene