[Quantization] Improve AWQ logic #19431
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
There was a problem hiding this comment.
Summary of Changes
Hello @jeejeelee, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request addresses a limitation in the AWQ quantization implementation by adding explicit support for Mixture-of-Experts (MoE) layers. It modifies the AWQ configuration logic to correctly identify and apply appropriate quantization methods for MoE modules, ensuring that models using MoE architectures can be quantized with AWQ.
Highlights
- MoE Support for AWQ: Added logic within the
AWQConfigto specifically handleFusedMoElayers during quantization method selection. - Quantization Method Selection for MoE: Implemented a mechanism to choose
AWQMoEMethodfor MoE layers if supported, falling back toMoeWNA16Config(configured with AWQ parameters) otherwise.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configureGemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request extends AWQ quantization to support Mixture-of-Experts (MoE) models when quantization="awq" is explicitly set. The AWQConfig.get_quant_method is updated to handle FusedMoE layers, with a fallback to MoeWNA16Config if AWQ Marlin MoE kernels are not suitable. A potential type mismatch when instantiating AWQMoEMethod is addressed by suggesting the construction of an AWQMarlinConfig to ensure type safety.
| elif isinstance(layer, FusedMoE): | ||
| # Lazy import to avoid circular import. | ||
| from .awq_marlin import AWQMoEMethod | ||
| from .moe_wna16 import MoeWNA16Config | ||
| from .utils.marlin_utils import check_moe_marlin_supports_layer | ||
| if not check_moe_marlin_supports_layer(layer, self.group_size): | ||
| logger.warning_once( | ||
| f"Layer '{prefix}' is not supported by AWQMoeMarlin. " | ||
| "Falling back to Moe WNA16 kernels.") | ||
| config = { | ||
| "quant_method": "awq", | ||
| "bits": self.weight_bits, | ||
| "group_size": self.group_size, | ||
| "zero_point": self.zero_point, | ||
| "lm_head": False, | ||
| } | ||
| return MoeWNA16Config.from_config(config).get_quant_method( | ||
| layer, prefix) | ||
| return AWQMoEMethod(layer.quant_config) |
There was a problem hiding this comment.
The AWQMoEMethod constructor expects an AWQMarlinConfig instance, but layer.quant_config is an AWQConfig instance when quantization="awq". Passing AWQConfig to AWQMoEMethod creates a type mismatch. Construct an AWQMarlinConfig instance using properties from the AWQConfig instance to ensure type safety.
from .awq_marlin import AWQMoEMethod, AWQMarlinConfig
# Construct an AWQMarlinConfig instance for type safety with AWQMoEMethod.
marlin_compatible_config_dict = {
"quant_method": "awq",
"bits": self.weight_bits,
"group_size": self.group_size,
"zero_point": self.zero_point,
"lm_head": False,
"modules_to_not_convert": self.modules_to_not_convert,
}
awq_marlin_config_for_moe = AWQMarlinConfig.from_config(marlin_compatible_config_dict)
return AWQMoEMethod(awq_marlin_config_for_moe)There was a problem hiding this comment.
Good catach, thanks!
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: minpeter <kali2005611@gmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: avigny <47987522+avigny@users.noreply.github.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.Purpose
Resolve the issue where AWQ does not support MoE when quantization is explicitly set
quantization=awqTest Plan
python examples/offline_inference/llm_engine_example.py
--model model_id #Arbitrary MoE AWQ model
--quantization awq
Test Result
(Optional) Documentation Update