[diffusion] refactor: separate runtime metadata from arch config

[mickqian]

Motivation

Modifications

Accuracy Tests

Speed Tests and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

[gemini-code-assist]

Code Review

This pull request refactors the multimodal model configuration system by migrating architectural fields from ArchConfig to ModelConfig and implementing a structured refresh mechanism for derived fields. It also introduces support for the LTX-2.3 model, including a new materialization overlay and specialized sampling parameter logic for two-stage pipelines. Review feedback highlights the need for a more robust check when detecting legacy post_init methods, recommends moving model-specific logic out of generic sampling utilities to prevent side effects, and suggests explicitly specifying UTF-8 encoding for file I/O in the new materialization script.

[gemini-code-assist]

The check type(arch_config).__dict__.get("__post_init__") is fragile because it only looks at the immediate class's dictionary. If a custom ArchConfig inherits its __post_init__ from a parent class (other than the base ArchConfig), it won't be detected here. A more robust check would be to compare the bound method's underlying function to the base implementation.

        if arch_config.__post_init__.__func__ is not ArchConfig.__post_init__:
            arch_config.__post_init__()
        else:
            arch_config.refresh_derived_fields()

[gemini-code-assist]

This block introduces model-specific logic into a generic utility method. It also modifies the sampling_params argument in-place, which is unexpected for a factory method and could lead to bugs if the default sampling_params instance is shared across requests. Consider moving this logic into a specialized method within LTX23SamplingParams or handling it during pipeline initialization.

[gemini-code-assist]

Specify encoding="utf-8" when opening files for reading or writing to ensure consistent behavior across different operating systems and environments.

Suggested change

	def _load_json(path: str) -> dict:
	with open(path) as f:
	return json.load(f)
	def _write_json(path: str, payload: dict) -> None:
	with open(path, "w") as f:
	json.dump(payload, f, indent=2)
	f.write("\n")
	def _load_json(path: str) -> dict:
	with open(path, encoding="utf-8") as f:
	return json.load(f)
	def _write_json(path: str, payload: dict) -> None:
	with open(path, "w", encoding="utf-8") as f:
	json.dump(payload, f, indent=2)
	f.write("\n")