[Bugfix] GLM-Image: fix noisy / washed-out t2i output (#3034)

[ptarasiewiczNV]

Summary

Fixes GLM-Image's noisy / washed-out t2i output reported in vllm-omni#3034. The recipe's minimal curl was producing a near-uniform white image (mean=249, std=15) instead of a coherent landscape.

Two independent fixes in vllm_omni/entrypoints/openai/serving_chat.py and vllm_omni/inputs/preprocess.py:

1. Route t2i requests through the multimodal processor.
OmniOpenAIServingChat only attached mm_processor_kwargs to the tprompt when the user supplied extra_body.height/width. OmniInputPreprocessor._process_text then gated its multimodal branch on elif mm_processor_kwargs: (truthiness), so when the field was omitted the default {} was falsy and routing fell back to plain _tokenize_prompt. That bypassed GLM-Image's HF processor and the image-generation scaffold <|image|>PROMPT<sop>H W<eop><sop>h w<eop><|dit_token_N|> it emits, so the AR never entered image-gen mode and collapsed to a handful of repeated VQ codes (unique=15/1281, no terminal EOS) which the DiT denoised into uniform white. Now serving_chat always attaches mm_processor_kwargs (possibly empty) for image-modality requests, and _process_text switches from truthiness to presence ("mm_processor_kwargs" in parsed_content) so an explicitly-empty dict correctly routes through the multimodal processor.

2. Make the AR max_tokens compute cover the default target size.
PR vllm-project#2320 dropped max_tokens: 1281 from the GLM-Image stage config and moved the compute into _apply_request_overrides, but gated it on height is not None and width is not None. For the bare-curl request (no extra_body) the gate skipped the compute and max_tokens fell through to max_model_len - seq_len (~131k), which produced the upstream IndexError that the original yaml edit was working around. Now when the user didn't pass h/w we fall back to any stage's default h/w (GLM-Image stage-1 yaml declares height: 1024, width: 1024), so the compute fires for the bare-curl too. The implicit gate becomes "a stage declares h/w in its sampling params" — LLM-only / audio pipelines skip, no architecture check needed. Also fixes a latent getattr(explicit_fields, "max_tokens", None) bug — explicit_fields is a set, so the getattr always returned None and silently overwrote user-provided max_tokens.

Before / after (same prompt, same seed=42)

	mean	std	min	max	unique AR codes	EOS emitted
Before	249	15	135	255	15 / 1281	❌
After	117	71	0	255	139 / 1281	✅

Second prompt sanity-check ("A red apple on a wooden table") also renders a coherent image (mean=117, std=82).

Test plan

• Reproduce the issue's bare curl: returns coherent landscape (1024×1024, mean=117, std=71).
• Second prompt ("A red apple on a wooden table") — coherent image.
• i2i smoke-test with an input image — no regression.
• pre-commit run --all-files.
• Existing GLM-Image unit tests still pass.

🤖 Generated with Claude Code