Bespoke Labs fine-tunes open models into repo-specific coding experts with SFT+GRPO

Bespoke Labs, with support from Thinking Machines, shows that post-training the open coding model Inkling with SFT+GRPO on a specific GitHub repo (fontTools) improves performance by 57 percentage points without sacrificing general capability.

2026-09-04 ~ 2026-09-04 · 4 related posts

• Bespoke Labs post-trains Inkling with SFT+GRPO, gaining 57pp on repo-level coding and 40% token efficiency — AlexGDimakis · 2026-09-04

3 near-duplicate retellings: AlexGDimakis · simonguozirui · heghbalz