Repository navigation
Restore the code-reading case that the snippet file was written for - #42
Merged
Merged
Conversation
`tools/bench/cases/snippets/lru_router.py` shipped with no case referencing it, so the inventory had no code-reading case even though the snippet exists for exactly that purpose. Add `code-review` on top of it. The task is deliberately narrowed to one defect plus one fixed method: the earlier broad "find every flaw and rewrite the class" wording made the model spend its whole output budget thinking and never emit code, which measures the thinking phase but not the coding phase. With the narrow wording a run produces visible code within 2048 tokens.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
tools/bench/cases/snippets/lru_router.pyshipped with no case referencing it, so the inventory had no code-reading case even though the snippet exists for exactly that purpose. This addscode-reviewon top of it.The task is narrowed to one defect plus one fixed method: the earlier broad "find every flaw and rewrite the class" wording made the model spend its entire output budget thinking and never emit code, which measures thinking but not coding. With the narrow wording a run produces visible code within 2048 tokens (verified: 459 content chars, 1010 reasoning tokens).
Measured on both engines over 3 repeats (medians): DeepInfer TTFT 0.469 s / TPOT 9.31 ms; the shared vLLM endpoint TTFT 0.484 s / TPOT 6.80 ms. Those runs are not a like-for-like engine comparison (different topology, shared endpoint), and no numbers are committed here.