ML teams still pin torch and Hugging Face deps to keep runs reproducible
vboykis · x · 2026-07-29
@vboykis asks how people manage ML dependencies in the face of supply-chain risk.
He notes that pinning versions in the torch / Hugging Face ecosystem is often the only practical way to keep training and inference runs reproducible, even though that makes Dependabot-style automation unhappy.
More from Infra
- Meta may pair higher capital spending with its first compute client announcement — RihardJarc · 2026-07-29
- Falling RAM prices could ease LLM infrastructure costs, as Micron and SK Hynix cool — CraftyPromise8304 · 2026-07-29
- Wayfinder adds Solana swaps beside EVM positions in Shell and SDK — templecrash · 2026-07-29
- A cloud platform exposes 46 MCP tools and asks how to gate destructive actions — Affectionate_Date749 · 2026-07-29
- Kimi K3 reportedly helped with kernel optimization and beats several models on in-house benches — nrehiew_ · 2026-07-29
- Kimi K3 report adds in-house coding, agent, and WebDev benchmark tables — nrehiew_ · 2026-07-29