Fix TorchAOConfig skip layers #19265

mobicham · 2025-06-06T10:00:03Z

[Follow-up to #19147 due to DCO rebasing issues]

Purpose

The goal of this small PR is to fix loading torchao models where not all the layers have been quantized.

The current implementation doesn't keep track of the skipped layers defined in config["modules_to_not_convert"]. As a result, quantized VL models where the vision head is not quantized results in a crash.

The PR also includes logic to skip layers defined in module_fqn_to_config. Currently, if a module is skipped in module_fqn_to_config, loading the model in vLLM would crash.

Also, made a quick fix to improve loading speed by avoiding creating an nn.Linear with the full tensor shape.

Test Plan

Dependencies

USE_CPP=0 pip install git+https://github.com/pytorch/ao -v --no-build-isolation --use-pep517;
pip install git+https://github.com/mobiusml/gemlite/;

Code

Loading a VL model with unquantized vision modules

import torch
from vllm import LLM
from vllm.sampling_params import SamplingParams

model_id = "mobiuslabsgmbh/Qwen2.5-VL-7B-Instruct_gemlite-ao_a16w4_gs_128_pack_32bit"
processor_args = {
    'limit_mm_per_prompt': {"image": 3}, 
    'mm_processor_kwargs': {"min_pixels": 28 * 28, "max_pixels": 1280 * 28 * 28},
    'disable_mm_preprocessor_cache': False,
}

llm = LLM(model=model_id, gpu_memory_utilization=0.9, dtype=torch.float16, max_model_len=4096, 
            max_num_batched_tokens=4096, **processor_args)

Skip module example

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TorchAoConfig

model_id = "meta-llama/Llama-3.1-8B-Instruct"

from torchao.quantization import Int4WeightOnlyConfig, Int8WeightOnlyConfig, ModuleFqnToConfig
config = Int4WeightOnlyConfig(group_size=128)
config2 = None 

quant_config = ModuleFqnToConfig({"_default": config, "model.layers.0.self_attn.o_proj": config2, "model.layers.13.self_attn.o_proj": config2})

quantization_config = TorchAoConfig(quant_type=quant_config)
quantized_model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", torch_dtype=torch.bfloat16, quantization_config=quantization_config)

print("quantized model:", quantized_model)
tokenizer = AutoTokenizer.from_pretrained(model_id)

q_model_id = 'quant_model_test'
quantized_model.save_pretrained(q_model_id, safe_serialization=False) 
tokenizer.save_pretrained(q_model_id)
######################################################################################

import torch
from vllm import LLM
from vllm.sampling_params import SamplingParams
llm = LLM(model="quant_model_test", dtype=torch.bfloat16) 
sampling_params = SamplingParams(max_tokens=1024, temperature=0.5, repetition_penalty=1.1, ignore_eos=False)
messages = [{"content": "You are a helpful assistant", "role":"system"}, {"content":"Solve this equation x^2 + 1 = -1.", "role":"user"}]
outputs = llm.chat(messages, sampling_params, chat_template=llm.get_tokenizer().chat_template)
print(outputs[0].outputs[0].text)

Test Result

The model should load successfully.

@jerryzh168

…atures) for faster loading Signed-off-by: mobicham <hicham@mobiuslabs.com>

Signed-off-by: mobicham <hicham@mobiuslabs.com>

gemini-code-assist

Hello @mobicham, 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!

Summary of Changes

Hello team,

Gemini here with a summary of this pull request. This PR, titled "Fix TorchAOConfig skip layers", aims to address issues encountered when loading models quantized with TorchAO where not all layers have been converted. Specifically, it fixes crashes that occur when loading models that define certain modules to be skipped from quantization, either via the modules_to_not_convert list or by setting their configuration to None in module_fqn_to_config. The PR ensures that vLLM correctly identifies and handles these unquantized layers during loading. Additionally, a minor optimization is included to improve the loading speed for TorchAO models by avoiding the creation of large dummy nn.Linear modules during parameter quantization.

Highlights

TorchAO Quantization Fix: Resolves issues loading TorchAO quantized models that contain layers explicitly marked to be skipped from quantization.
Support for Skipped Modules: Adds logic to correctly handle modules listed in modules_to_not_convert and modules with a None configuration in module_fqn_to_config, ensuring they are treated as unquantized.
Loading Speed Optimization: Improves the speed of loading TorchAO models by optimizing the creation of dummy nn.Linear modules used during parameter quantization.
New Test Case: Adds a test case specifically for loading a Qwen-VL model quantized with TorchAO, which helps validate the fix for models with unquantized components like vision heads.

Changelog

tests/quantization/test_torchao.py
- Added a new test function test_qwenvl_int8wo_model_loading_with_params (lines 62-74) to test loading a Qwen-VL model quantized with TorchAO, targeting the scenario with potentially skipped layers.
vllm/model_executor/layers/quantization/torchao.py
- Modified the TorchAOConfig constructor to accept an optional skip_modules list (lines 23-25, 41).
- Updated the from_config class method to parse modules_to_not_convert and identify modules set to None in module_fqn_to_config, adding them to the skip_modules list (lines 78-85).
- In get_quant_method, added a check to return UnquantizedLinearMethod() if the current module's prefix is in the skip_modules list (lines 96-97).
- Ensured the skip_modules list is passed down when creating nested TorchAOConfig instances within get_quant_method (line 105).
- Optimized torchao_quantize_param_data by creating a small nn.Linear(1, 1) and manually setting in_features and out_features instead of using the full parameter shape directly (lines 129-131).

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.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request effectively addresses the issue of loading TorchAO models with partially quantized layers, particularly for Vision-Language models. The changes to handle modules_to_not_convert and module_fqn_to_config for skipping layers are well-implemented. The added test case for a Qwen-VL model is a good addition, and the optimization in torchao_quantize_param_data to reduce memory allocation during dummy linear layer creation is a nice improvement.

I have one suggestion regarding the string matching logic for skipping modules, which could be made more robust to prevent potential over-matching. Overall, this is a valuable fix.

Summary of Findings

Module Skipping Logic Robustness: The logic for determining whether to skip a module (any(s in prefix for s in self.skip_modules)) uses a general substring check. This could potentially lead to over-matching if a skip pattern is a substring of an unrelated module's FQN (e.g., skipping "layer.1" might unintentionally affect "layer.10"). A more precise FQN-aware prefix matching or exact matching would be more robust.
Test Coverage: A new test case (test_qwenvl_int8wo_model_loading_with_params) was added, which is good for verifying the fix for VL models with unquantized vision modules.
Performance Improvement: The change in torchao_quantize_param_data to initialize nn.Linear with minimal dimensions (1,1) and then update in_features and out_features is a good optimization to reduce temporary memory allocation.

Merge Readiness

The pull request is well-structured and addresses the core issues effectively. However, there's one medium-severity concern regarding the robustness of the module skipping logic that should be discussed and potentially addressed. Once that point is clarified or resolved, the PR should be in good shape for merging. As an AI, I am not authorized to approve pull requests; please ensure further review and approval from the maintainers.

vllm/model_executor/layers/quantization/torchao.py

Signed-off-by: mobicham <hicham@mobiuslabs.com>

github-actions · 2025-06-06T10:21:26Z

👋 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 fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

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 ready label to the PR or enable auto-merge.

🚀

Signed-off-by: mobicham <hicham@mobiuslabs.com>

houseroad · 2025-06-06T13:13:00Z

cc: @jerryzh168

jerryzh168 · 2025-06-06T14:27:24Z

@mobicham thanks for the fix, can you talk a bit more about qkv fusion that you mentioned before? still didn't quite get it

vllm/model_executor/layers/quantization/torchao.py

drisspg

The q,k,v fusion issue you mentioned makes sense, does this PR fix that?

mobicham · 2025-06-07T08:57:43Z

@drisspg only if:

They have the same module-level quant settings.
The ao implementation correctly implements slice and copy_.

Otherwise, there's no clean way to merge qkv if they don't have the quant settings. Moreover, the merging is not happening in TorchAOConfig, it's happening in the QKV linear modules.

The main focus of this PR is to handle layer skipping for layers that were not quantized though. So it's simply checking in the config if the prefix matches the skipped layers defined in the config.

drisspg · 2025-06-07T13:54:32Z

Ohhh yea, fwiw I am planning on writing a little doc in AO on how to get write a subclass that will work with VLLM, and the slice and copy is the main point.

I guess it feels like if someone skipped one of the q,k, projections then we shud skip (via ModFQNconfig or skip list) the stacked variant. Was more curious if this is tested and expected to work in this PR

mobicham · 2025-06-07T14:05:56Z

I guess it feels like if someone skipped one of the q,k, projections then we shud skip (via ModFQNconfig or skip list) the stacked variant. Was more curious if this is tested and expected to work in this PR

I see! That can't work with the current code unfortunately because the metadata will mismatch during the slice/copy.
However, maybe it's worth mentioning it in the doc or throw a warning if q,k,v don't share the same config. Same thing for the MLP layer by the way.

houseroad · 2025-06-08T08:08:04Z

Actually DCO is only a soft requirement. :-)

vllm/model_executor/layers/quantization/torchao.py

Signed-off-by: mobicham <hicham@mobiuslabs.com>

mobicham · 2025-06-11T09:24:00Z

Anything else to have this merged? Thank you!

houseroad

Could you rebase to the latest main and fix the pre-commit linter?

Signed-off-by: mobicham <hicham@mobiuslabs.com>

mobicham · 2025-06-12T08:30:17Z

@houseroad sorry didn't see that, fixed, thank you!

Signed-off-by: mobicham <hicham@mobiuslabs.com> Signed-off-by: minpeter <kali2005611@gmail.com>

Signed-off-by: mobicham <hicham@mobiuslabs.com>

Signed-off-by: mobicham <hicham@mobiuslabs.com> Signed-off-by: avigny <47987522+avigny@users.noreply.github.com>

Signed-off-by: mobicham <hicham@mobiuslabs.com>

mobicham added 2 commits June 6, 2025 09:43

Fix layer skip in TorchAOConfig + avoid nn.Linear(in_features, out_fe…

5097538

…atures) for faster loading Signed-off-by: mobicham <hicham@mobiuslabs.com>

handle undefined/None module_fqn_to_config

b85e085

Signed-off-by: mobicham <hicham@mobiuslabs.com>

mobicham requested review from mgoin, robertgshaw2-redhat and tlrmchlsmth as code owners June 6, 2025 10:00

gemini-code-assist bot reviewed Jun 6, 2025

View reviewed changes

mobicham mentioned this pull request Jun 6, 2025

Fix TorchAOConfig skip layers #19147

Closed

gemini-code-assist bot suggested changes Jun 6, 2025

View reviewed changes

vllm/model_executor/layers/quantization/torchao.py Outdated Show resolved Hide resolved

prefix/skip_modules match logic

98dae52

Signed-off-by: mobicham <hicham@mobiuslabs.com>

mobicham added 4 commits June 6, 2025 11:36

prefix/skip_modules match logic

ce54576

Signed-off-by: mobicham <hicham@mobiuslabs.com>

Merge branch 'main' of https://github.com/mobicham/vllm

db8f946

pre-commit fix

78f40d5

Signed-off-by: mobicham <hicham@mobiuslabs.com>

pre-commit fix

ba98592

Signed-off-by: mobicham <hicham@mobiuslabs.com>

drisspg reviewed Jun 6, 2025

View reviewed changes

vllm/model_executor/layers/quantization/torchao.py Outdated Show resolved Hide resolved

drisspg reviewed Jun 6, 2025

View reviewed changes

mgoin reviewed Jun 9, 2025

View reviewed changes

vllm/model_executor/layers/quantization/torchao.py Outdated Show resolved Hide resolved

add minimal nn Linear alloc comment

f7f8578

Signed-off-by: mobicham <hicham@mobiuslabs.com>

mgoin added bug Something isn't working quantization ready ONLY add when PR is ready to merge/full CI is needed labels Jun 9, 2025

use meta device to init dummy linear

0a0b1a8

Signed-off-by: mobicham <hicham@mobiuslabs.com>

drisspg approved these changes Jun 10, 2025

View reviewed changes

houseroad reviewed Jun 11, 2025

View reviewed changes

mobicham and others added 3 commits June 12, 2025 08:03

pre-commit fix

eae62b2

Signed-off-by: mobicham <hicham@mobiuslabs.com>

Merge branch 'vllm-project:main' into main

d920cfa

pre-commit fix

ba7bf5c

Signed-off-by: mobicham <hicham@mobiuslabs.com>

houseroad approved these changes Jun 12, 2025

View reviewed changes

houseroad merged commit 96846bb into vllm-project:main Jun 12, 2025
71 checks passed

mobicham mentioned this pull request Jun 12, 2025

vLLM api server patching mobiusml/hqq#152

Closed

minpeter pushed a commit to minpeter/vllm that referenced this pull request Jun 24, 2025

Fix TorchAOConfig skip layers (vllm-project#19265)

07e7b09

Signed-off-by: mobicham <hicham@mobiuslabs.com> Signed-off-by: minpeter <kali2005611@gmail.com>

xjpang pushed a commit to xjpang/vllm that referenced this pull request Jun 30, 2025

Fix TorchAOConfig skip layers (vllm-project#19265)

3c8ff80

Signed-off-by: mobicham <hicham@mobiuslabs.com>

wseaton pushed a commit to wseaton/vllm that referenced this pull request Jun 30, 2025

Fix TorchAOConfig skip layers (vllm-project#19265)

de91cba

Signed-off-by: mobicham <hicham@mobiuslabs.com>

avigny pushed a commit to avigny/vllm that referenced this pull request Jul 31, 2025

Fix TorchAOConfig skip layers (vllm-project#19265)

2ba5fe9

Signed-off-by: mobicham <hicham@mobiuslabs.com> Signed-off-by: avigny <47987522+avigny@users.noreply.github.com>

googlercolin pushed a commit to googlercolin/vllm that referenced this pull request Aug 29, 2025

Fix TorchAOConfig skip layers (vllm-project#19265)

0c33efc

Signed-off-by: mobicham <hicham@mobiuslabs.com>

Uh oh!

Fix TorchAOConfig skip layers #19265

Fix TorchAOConfig skip layers #19265

Uh oh!

Conversation

mobicham commented Jun 6, 2025 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Test Plan

Dependencies

Code

Loading a VL model with unquantized vision modules

Skip module example

Test Result

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Summary of Changes

Highlights

Changelog

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Summary of Findings

Merge Readiness

Uh oh!

Uh oh!

github-actions bot commented Jun 6, 2025

Uh oh!

houseroad commented Jun 6, 2025

Uh oh!

jerryzh168 commented Jun 6, 2025

Uh oh!

Uh oh!

drisspg left a comment

Choose a reason for hiding this comment

Uh oh!

mobicham commented Jun 7, 2025

Uh oh!

drisspg commented Jun 7, 2025

Uh oh!

mobicham commented Jun 7, 2025

Uh oh!

houseroad commented Jun 8, 2025

Uh oh!

Uh oh!

mobicham commented Jun 11, 2025

Uh oh!

houseroad left a comment

Choose a reason for hiding this comment

Uh oh!

mobicham commented Jun 12, 2025

Uh oh!

Uh oh!

Uh oh!

mobicham commented Jun 6, 2025 •

edited by github-actions bot

Loading