Reduce VRAM usage of quantizing VLM models#1777
Merged
Merged
Conversation
Signed-off-by: lvliang-intel <liang1.lv@intel.com>
for more information, see https://pre-commit.ci
Contributor
There was a problem hiding this comment.
Pull request overview
This PR targets issue #1744 by enabling a lower-VRAM calibration/caching path when quantizing VLM/MLLM models (notably in the new-architecture compressor), and adds a CUDA test intended to validate reduced peak VRAM for a text-only MLLM calibration flow.
Changes:
- Allow
compressors_newCPU caching path for MLLM models by removing the explicit MLLM exclusion intry_cache_inter_data_gpucpu. - Add a CUDA test that runs MLLM quantization with a text dataset and asserts peak VRAM stays low.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.
| File | Description |
|---|---|
auto_round/compressors_new/calib.py |
Removes the MLLM-specific guard so CPU caching can apply to MLLM in the new compressor path. |
test/test_cuda/models/test_mllm.py |
Adds a low-VRAM regression test for MLLM quantization with a list-of-strings dataset using memory_monitor. |
Contributor
Author
|
/azp run Unit-Test-CUDA-AutoRound |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Contributor
Author
|
/azp run Unit-Test-CUDA-AutoRound |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Contributor
Author
|
/azp run Unit-Test-CUDA-AutoRound |
|
Azure Pipelines successfully started running 1 pipeline(s). |
…ix_vlm_large_vram Signed-off-by: lvliang-intel <liang1.lv@intel.com>
Signed-off-by: lvliang-intel <liang1.lv@intel.com>
Contributor
Author
|
/azp run Unit-Test-CUDA-AutoRound |
|
Azure Pipelines successfully started running 1 pipeline(s). |
n1ck-guo
approved these changes
May 20, 2026
This file contains hidden or 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
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.
Description
Fix issue #1744, reduce VRAM usage of quantizing VLM models.
CUDA_VISIBLE_DEVICES=0 python -m auto_round /home/lvl/models/Qwen3-VL-8B-Instruct --bits 4 --group_size 128 --dataset "pile-10k" --nsamples 4 --seqlen 512 --iters 2 --output_dir ./tmp_vlm_quant
Type of Change
Bug fix
Related Issues
#1744
Checklist Before Submitting
/azp run Unit-Test-CUDA-AutoRound.