Spaces:

minhtudragon
/

headroom_2

Build error

tudragon154203 Claude Happy commited on Jun 15

Commit

6b0588d

1 Parent(s): 5945d36

evals: fix conclusion accuracy claim to match Pass 2 ground_truth results

- Updated '62.5pp' → '5.0pp' to reflect actual 12.5% → 17.5% gain

Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>

Files changed (1) hide show

evals/REPORT-minimax-m3-2026-06-15.md +1 -1

evals/REPORT-minimax-m3-2026-06-15.md CHANGED Viewed

@@ -51,7 +51,7 @@ minimax-m3 chokes on long noisy context. Headroom strips noise → model focuses
 ## 3. Conclusion
-- **Long context: huge win.** 22% token savings + 62.5pp accuracy gain. No regressions.
 - **LCB codegen: mixed.** Token savings real (38.8%) but 1 real regression traced to user-message compression. Disabling that flag = noise floor only.
 - **Recommendation:** Enable headroom for long-context / tool-output / RAG workloads. Keep `compress_user_messages=False` for codegen prompts where the problem statement must stay verbatim.

 ## 3. Conclusion
+- **Long context: modest win.** 22% token savings + 5.0pp accuracy gain. No regressions.
 - **LCB codegen: mixed.** Token savings real (38.8%) but 1 real regression traced to user-message compression. Disabling that flag = noise floor only.
 - **Recommendation:** Enable headroom for long-context / tool-output / RAG workloads. Keep `compress_user_messages=False` for codegen prompts where the problem statement must stay verbatim.