[AutoRound] Support GLM-Image W4A16 quantization model

[lvliang-intel]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Support GLM-Image W4A16 AutoRound quantization in vLLM-Omni, extending the existing AutoRound W4A16 infrastructure (originally built for FLUX and Qwen3-Omni) to the GLM-Image diffusion model. This reduces model size by ~4x and GPU memory footprint while preserving generation quality.

https://huggingface.co/Intel/GLM-Image-int4-AutoRound

Related: #1325, #1777, #2670

Key changes:
Replace all nn.Linear / ColumnParallelLinear / RowParallelLinear projection layers in the GLM-Image DiT with their vLLM quantized-aware counterparts (ReplicatedLinear, ColumnParallelLinear, RowParallelLinear with quant_config).
Also added contiguous calls before RowParallelLinear (required for FP8/W4A16 kernels) and tuple-unpacking for ReplicatedLinear output.

Test Plan

E2E offline inference tests added.
TIIF-Bench accuracy evaluation test.
DPG-Bench accuracy evaluation test.

Test Result

TIIF-Bench Accuracy (9 Sub-Attributes Average)

Model	overall-short	overall-long
glm-image-ar-w4a16	0.8175	0.8645
glm-image (BF16 baseline)	0.8277	0.8903

Summary:

• W4A16 quantized model retains:
  • 98.8% (short)
  • 97.1% (long)
    Average accuracy drop: ~1.3% Accuracy degradation is minimal and within an acceptable range for 4-bit quantization.

---

Model Size Reduction

Component	BF16 Baseline	W4A16 AutoRound	Reduction
Total	~34 GB	~13 GB	~62%

Overall checkpoint is ~3.8× smaller

---

E2E Generation Smoke Test

• ✅ Text-to-Image: Functional
• ✅ Image-to-Image: Functional
• ✅ Output Quality: Valid, non-blank images
• ✅ Resolution: 256 × 256

The quantized W4A16 model maintains full pipeline functionality with no critical degradation in generation behavior.

---

Performance Test Result on A100

Metric	BF16 (Original)	W4A16 (AutoRound)	Δ
Latency Mean	60.21 s	57.98 s	-3.7%
Latency Median	60.21 s	57.96 s	-3.7%
Latency P50	60.21 s	57.96 s	-3.7%
Latency P95	60.38 s	58.20 s	-3.6%
Latency P99	60.43 s	58.26 s	-3.6%
Throughput	0.0166 qps	0.0172 qps	+3.8%
Peak Memory	23294 MB (22.8 GiB)	13680 MB (13.4 GiB)	-41.3%
Requests	64	64	—
Duration	3853 s	3711 s	-3.7%
Failed Requests	0	0	—

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
• (Optional) Release notes update. If your change is user-facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[hsliuustc0106]

BLOCKING:

• Documentation — AutoRound documentation table should be updated. Please add GLM-Image to the supported models table in docs/user_guide/diffusion/quantization/autoround.md:

  | GLM-Image | Intel/GLM-Image-int4-AutoRound | W4A16 | 128 | GPTQ-Marlin |

[hsliuustc0106]

please add the latency test results as well

[lvliang-intel]

please add the latency test results as well

Will update the performance test result soon.

[lvliang-intel]

BLOCKING:

• Documentation — AutoRound documentation table should be updated. Please add GLM-Image to the supported models table in docs/user_guide/diffusion/quantization/autoround.md:
  | GLM-Image | Intel/GLM-Image-int4-AutoRound | W4A16 | 128 | GPTQ-Marlin |

Thanks for reminding me this. Doc updated.

[zhumingjue138]

Please add the necessary ut test cases.

[lvliang-intel]

Please add the necessary ut test cases.

ut added.

[lishunyang12]

Can you try with longer seq？

[lvliang-intel]

Can you try with longer seq？

Sure, I will run the performance test with longer sequence.

[david6666666]

Merge conflicts need fixing before review.