yashash045/devops-pipeline-gym-sft-adapter

Update README.md: humanise voice, narrative, captions

README: restructure table to judgment_call like-for-like (was mixing all-task aggregates)

Honest reframe: baseline is Qwen2.5-7B (not Qwen3-1.7B)

Model card: add concrete -0.044 / +1.156 number (was placeholder pointing at BLOG)

Replace auto-generated card with proper writeup + frontier comparison + links

Kaggle frontier eval: gpt_oss_120b

Kaggle frontier eval: mistral_large

Kaggle frontier eval: deepseek_v31

Kaggle frontier eval: llama33_70b

Kaggle frontier eval: qwen25_72b

Phase M Stage 1: SFT warmup adapter

initial commit