e2e: survive malformed tool-call JSON; record small-model results#117
Merged
Conversation
llama-3.1-8b (3 rounds, all unsound) exposed a harness crash: invalid JSON in tool-call arguments raised JSONDecodeError mid-run. Such calls now get an error tool-result so the model can retry, and the README gains an observed-results table (qwen-72b sound on round 2 vs llama-8b's three distinct failure modes) — the weak-model thesis, measured. Co-Authored-By: Claude Fable 5 <noreply@anthropic.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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follow-up to #113/#116, from testing the smaller model as discussed.
What happened
Three rounds with
meta-llama/llama-3.1-8b-instruct(the suggested small model), all unsound, each failing differently:Why it's worth recording
It sharpens the doc's thesis with data: qwen-72b needed the fix loop and used it (sound on round 2); llama-8b can't sustain the loop at all — and the soundness gate correctly rejected every round. The e2e README now carries an observed-results table.
Changes
harness.py: malformed tool-call JSON → error tool-result instead of a crash (--dry-runstill green)e2e/llm/README.md: observed-results tableTranscripts of the failed rounds stay local/gitignored as usual.
🤖 Generated with Claude Code