refactor(inference): data-driven config exclusion

[adibarra]

Note

Medium Risk
Touches inference legend selection, presets, and DSv4 default chart behavior; behavior is intended to match prior MTP rules plus new ATOM/SGLang grouping, with broad test coverage but user-visible chart filtering changes.

Overview
Replaces the hard-coded MTP engine-family exclusion (mtp-exclusion.ts + mtpEngineExclusion flag) with a data-driven exclusion module: models declare ExclusionSpec[] (suffix, prefix stripping, group aliases), compiled via buildExclusion for legend toggles, presets, and auto-reset.

DeepSeek V4 Pro now uses MTP_ENGINE_EXCLUSION on *_mtp keys, with ATOM and SGLang in one comparability group (groupAliases: { atom: 'sglang' }) so they can appear together but stay mutually exclusive with vLLM MTP. Preset bare-GPU matching skips keys ending in any configured exclusion suffix via getModelExclusion.

InferenceContext wires the generic helpers (resolveExclusionToggle, clearAllExclusionGroups, effectiveLegendItems); existing MTP conflict toasts are unchanged. Tests move from mtp-exclusion.test.ts to exclusion.test.ts with coverage for the ATOM/SGLang group behavior.

^{Reviewed by Cursor Bugbot for commit e28bb0d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
inferencemax-app	Ready	Preview, Comment	Jun 2, 2026 12:34am

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit e28bb0d. Configure here.}