Qwen Decision 27B — Decision Index 0.3 results
0
10 commits
2 linked in READMEs
updated Oct 9, 2026
Complete public evaluation of Qwen Decision 27B noncausal LoRA, selected step1533. See runs/noncausal-step1533/scores.json: complete: true, 140178 scoreable requests, public index 57.70.
The run directory contains untouched compact predictions, official aggregate reports, environment metadata and the exact evaluated server. It does not contain benchmark prompts or the source corpus. Rebuild the corpus using the official kit at commit 9eb2dbe2a358004c8782c66e40a83ac07b953fec.
This is a public-only score, not an official Full leaderboard score. Maintainers independently rescore results, evaluate private tasks and measure latency. Training/public-benchmark overlap has not been fully audited; no contamination-free claim is made. See the model card for full training and capacity disclosure.
The run resumed 58142 previously saved predictions with the same checkpoint and inference protocol. It processed 140620 total requests, including442 excluded by official scoring; the 140178 scoreable set has140148 supported and30 unsupported responses, with no execution errors.
Qwen Decision 27B — Decision Index 0.3 results
0
10 commits
2 linked in READMEs
updated Oct 9, 2026
Complete public evaluation of Qwen Decision 27B noncausal LoRA, selected step1533. See runs/noncausal-step1533/scores.json: complete: true, 140178 scoreable requests, public index 57.70.
The run directory contains untouched compact predictions, official aggregate reports, environment metadata and the exact evaluated server. It does not contain benchmark prompts or the source corpus. Rebuild the corpus using the official kit at commit 9eb2dbe2a358004c8782c66e40a83ac07b953fec.
This is a public-only score, not an official Full leaderboard score. Maintainers independently rescore results, evaluate private tasks and measure latency. Training/public-benchmark overlap has not been fully audited; no contamination-free claim is made. See the model card for full training and capacity disclosure.
The run resumed 58142 previously saved predictions with the same checkpoint and inference protocol. It processed 140620 total requests, including442 excluded by official scoring; the 140178 scoreable set has140148 supported and30 unsupported responses, with no execution errors.
