40Hz
/

autoresearch-coding-v1

Generated from Trainer

Model card Files Files and versions

Metrics Training metrics Community

autoresearch-coding-v1 / evals

47.7 kB

Ctrl+K

Ctrl+K

2 contributors

History: 4 commits

40Hz's picture

Gate 1 PASS: coding pass@1 = 0.95 (adapter) vs 0.22 (base), greedy 1.0, 95/100

1a7362d verified about 2 months ago

IMPROVEMENTS.md

4.36 kB
Add eval harness v2 (pass@k, per-class mAP, nDCG, leakage check) + improvements doc about 2 months ago
colab_run_all.py

10.7 kB
add colab T4 driver for all three tracks about 2 months ago
eval_business.py

6.09 kB
Add eval harness v2 (pass@k, per-class mAP, nDCG, leakage check) + improvements doc about 2 months ago
eval_coding.py

7.92 kB
Add eval harness v2 (pass@k, per-class mAP, nDCG, leakage check) + improvements doc about 2 months ago
eval_ui_detr.py

7.07 kB
Add eval harness v2 (pass@k, per-class mAP, nDCG, leakage check) + improvements doc about 2 months ago
gate1_colab_eval.py

8.28 kB
Gate 1 eval: add fixed extraction (markdown-fence + signature restatement) and torchao pin; adapter loads correctly about 2 months ago
gate1_result.json

783 Bytes
Gate 1 PASS: coding pass@1 = 0.95 (adapter) vs 0.22 (base), greedy 1.0, 95/100 about 2 months ago
leakage_check.py

2.51 kB
Add eval harness v2 (pass@k, per-class mAP, nDCG, leakage check) + improvements doc about 2 months ago