Developer asks how to turn real codebases into instruction-to-code fine-tuning datasets
ImBadGuyInEveryStory · reddit · 2026-09-11
A Reddit poster wants to convert several of their well-designed web projects (React/Next.js components, pages, styling, or static HTML sites) into instruction/prompt → code fine-tuning datasets for training instruct/thinking/diffusion coding models.
Key open questions they raise:
- Preserving context across components and files in training samples
- Pairing screenshots with code
- Generating meaningful instructions instead of generic descriptions
They also mention designing a new model architecture aimed at better quality/speed with lower VRAM, and want a solid dataset and benchmark to evaluate it properly, asking for tools, repos, papers, and workflows.
More from coding & agent
- Why one agent instance must serve one run: lessons from smolagents source — Mahmoud_Zalt · 2026-09-11
- Prompting won't guarantee pure JSON: why teams use grammar-guided decoding — dotey · 2026-09-11
- supermemory kills company/personal brain products to focus on agent memory API — julianweisser · 2026-09-11
- Two AI agents ping-pong refund emails back and forth in seconds — robleclerc · 2026-09-11
- visual-explainer: Agent Skill Turns Terminal Output Into Styled HTML, 9.7k GitHub Stars — tom_doerr · 2026-09-11
- One prompt to check if your paid AI got nerfed: an SVG pelican on a bike — lxfater · 2026-09-11