Nightly 3034 (Simple · Model Executor Test) failed all six
test_generation_prompt_ids_preserve_hf_target_scaffold cases with
AttributeError: type object 'GlmImageProcessor' has no attribute '_build_prompt_with_target_shape'
The 0.29 rewrite of GlmImageMultiModalProcessor.apply reached for
GlmImageProcessor._build_prompt_with_target_shape, a private transformers
method. The test suite stubs transformers.models.glm_image.processing_glm_image
with a bare class to import glm_image_ar without the real package, so in the
full CPU run the private attribute is not there; the same dependency would
break at runtime on any transformers build that renames or removes it.
Format the scaffold in Omni instead: _build_target_shape_scaffold mirrors the
HF formula (identical in transformers 5.13 and 5.14: target grid, plus the
preview grid for text-to-image, then the image bos token) using only the
processor's public grid_bos_token / grid_eos_token / bos_token attributes.
The test keeps its hard-coded expected suffix, which is what HF produces for
512x768, and no longer imports the HF class.
Validation: tests/model_executor/models/glm_image passes (25 tests) both alone
and inside the full `tests/model_executor -m "core_model and cpu"` run.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MpXe2LeRrUQGjZfKFaocMZ
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
Reuse upstream weight mapping and multimodal processing, consolidate Omni processor bridges, and trim unused Hunyuan internals. Route no-media kwargs through the active renderer and preserve GLM generation prompt formatting.
Validation: 453 CPU tests passed, 7 skipped; processor imports, Ruff, and diff checks passed. GPU and end-to-end CI not rerun.
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
