Automated MNLP evaluation report (2026-06-10) (#4)

Automated MNLP evaluation report (2026-06-09) (#3)

SFT v2 + rep 1.03, sampling, max_new 16384 (restore CI 0.44 config)

RLVR v2 r1 ckpt-1875 + rep_penalty 1.04

Update generation_config.json

Push SFT v2 best (rep_penalty 1.02, sampling, max_new 16000)

noth + rlvr chckpt1000 greedy

SFT v3 noTh

sft v2 best perf CI yet

Phase 4 push: merged Qwen3-1.7B + LoRA adapter from final/ (fp16). max_new_tokens=4096.

Upload folder using huggingface_hub

SFT v2 + repetition_penalty 1.03 (revert vers ~rep 1.0)

SFT v2 + repetition_penalty 1.2 in generation_config (internal no_box ~14% -> 0.7%)

RLVR v2 r1 ckpt-1875

Automated MNLP evaluation report (2026-05-18) (#2)

Phase 4 push: merged Qwen3-1.7B + LoRA adapter from best/ (fp16). max_new_tokens=4096.

Automated MNLP evaluation report (2026-05-13) (#1)

RLVR v3 ckpt-100 (full val pass@1=0.6189, +3.38pp / 2.5σ vs SFT)

Phase 4 push: merged Qwen3-1.7B + LoRA adapter from final/ (fp16). max_new_tokens=4096.

initial commit