feat(benchmark): separate judge VLM from pipeline VLM#253
Open
dippatel1994 wants to merge 1 commit into
Open
Conversation
Add --judge-provider/--judge-model (JUDGE_VLM_PROVIDER/JUDGE_VLM_MODEL) so the benchmark judge can use a different, stronger model than the generation pipeline. create_vlm() gains a model_override param that takes precedence over provider-specific and generic model fields. The benchmark judge factory swaps provider/model when configured, leaving pipeline settings untouched. Covered by tests for fallback and override behavior.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Adds the ability to run the benchmark judge on a different model than the generation pipeline:
--judge-provider/--judge-modelonpaperbanana benchmark.judge_vlm_provider/judge_vlm_model(env:JUDGE_VLM_PROVIDER/JUDGE_VLM_MODEL, plusjudge.*YAML keys).ProviderRegistry.create_vlm()gains amodel_overrideparam that takes precedence over provider-specific and generic model fields, without mutating the passed settings.Why
The judge and the generation pipeline previously shared one VLM, so
--vlm-providerchanged both. Separating them lets you pair a cheap pipeline model with a stronger, independent judge (and keeps the judge fixed when comparing pipeline variants).Tests
New coverage in
test_benchmark.py(judge defaults to pipeline VLM; uses judge model/provider override; pipeline settings unchanged) andtest_features.py(model_overrideprecedence, no settings mutation). Full affected suites green (66 passed).Backwards compatible: both new settings default to
None, preserving existing behavior.