Bespoke Labs post-trains models on code repos with SFT plus GRPO
AlexGDimakis · x · 2026-09-04
New research from Bespoke Labs (with support from Thinking Machines / John Schulman): post-training a model to specialize on a given GitHub repository. Starting from the Inkling base, they run SFT on trajectories from a strong teacher model, then GRPO reinforcement learning on repo-specialized curated environments, substantially improving repo-specific performance.
Related event: Bespoke Labs Trains Models into Repo-Specific Coding Experts(2 posts)→
More from coding & agent
- xAI explains Grok Bot: designing UI for persistent, self-starting agents — soleio · 2026-09-04
- youtube-skills: open-source toolkit giving AI agents YouTube transcripts and search — tom_doerr · 2026-09-04
- Developer amazed by free local model Muse Spark 1.3 running fast in OpenCode — HumungreousNobolatis · 2026-09-04
- Go team ships goroutine leak profiler that detects blocked channels in production — rseroter · 2026-09-04
- Catch AI launches a $99/mo executive admin agent that proactively saves $400 on hotels — eyishazyer · 2026-09-04
- Agent Token Breakdown: Only 50 of 1,300 Tokens Are Orchestration, Cutting Incident Cost to $0.001 — kalyan_kpl · 2026-09-04