Repository navigation
Conversation
The non-handoff PP completion path classifies the first token but drops the result when sending Note::Finish. A Stop terminal without a typed cause asks the frontend bridge to append its legacy compatibility EOS, although the stopping token has already been emitted. Keep the classification in the completion notes and preserve it through Front and the scheduler ledger. Factor only this emit/finish pair into a private helper so the regression exercises production completion logic without a model or CUDA runtime. Cover primary EOS precedence, explicit stop tokens, ignored EOS, one-token length finishes, token accounting, and absence of output after completion. Linux CPU regression fails before at the missing typed cause and passes after; all 48 serving tests, formatting and all-target kern-serve Clippy with warnings denied pass. The HTTP bridge is source-audited, not executed by these tests; no HTTP end-to-end, model or throughput claim is made. Signed-off-by: Venkat Raman <vraman2811@gmail.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The PP non-handoff path classifies its first output token but discards the typed cause when emitting completion. Preserve that cause through
Frontand the scheduler ledger so EOS and explicit-stop finishes remain distinguishable from length finishes.The model-free Linux regression exercises the production completion helper,
Finish::notes,Frontand the ledger. It compiles and fails before atstop_cause: Noneinstead ofSome(Eos(2)), then passes after. It also covers explicit stops, ignored EOS, ordinary length completion, token counts and no output after completion. All 48kern-servetests, formatting and all-target Clippy with warnings denied pass.The HTTP bridge was source-audited: a Stop finish without a cause takes its compatibility-EOS fallback. These tests do not execute that HTTP bridge or establish HTTP end-to-end accounting; no model or throughput claim is made. Dependencies and the lockfile are unchanged.