ML engineer: writing a bs=1 trainer myself beats explaining it to an LLM
cephaloform · x · 2026-10-01
User cephaloform argues that implementing a bs=1 version of a trainer is easier than explaining how to do it to an LLM — tacit engineering intuition LLMs can't yet absorb. In the follow-up he offers the flip side: agents shine at parallelizing whatever wacky bs=1 trainer you come up with, which used to be the most annoying part of multi-component training setups. A crisp take on where LLMs help and fail in ML engineering.
Related event: Developer: Build a bs=1 Trainer and Let AI Parallelize It(4 posts)→
More from coding & agent
- Runway MCP Lands in ChatGPT App Directory: Generate Video From Any Agent — tlakomy · 2026-10-01
- ComfyUI test: Qwen Image 2.1 beats FLUX.2 Krea Turbo at character sheets, ~3x faster — cgpixel23 · 2026-10-01
- Developer uses Claude Opus to turn hundreds of Three.js experiments into a music video — creatoroff · 2026-10-01
- Agent security mindset: least privilege to monitoring, in five layers — goyalshaliniuk · 2026-10-01
- 5 AI agent security risks: browsing, APIs and new attack surfaces — goyalshaliniuk · 2026-10-01
- Dev: I'd rather write dumb utils to parallelize torch trainers than use LLMs for the code — cephaloform · 2026-10-01