← sources

If open weights dry up, what are you archiving?

source: reddit (reddit:LocalLLaMA)published 2026-11-13 00:27ingested 2026-11-14 02:42

retained text

With Jensen out at NVIDIA, the "local AI is dangerous" narrative picking up steam, the possible ban on Chinese models, and hosted models locking down RE/cyber work, I'm starting to think about a worst case where Chinese weights get banned and US labs stop releasing open models. So what's your plan?

**What models are you archiving?** Just the biggest/best, or a spread?

**Which ones for which jobs?** Coding, RE/security, vision, embeddings, RAG, small fast models for agents, etc.

**Training tools:** if the last good big model we get is something like Kimi K3, what are you keeping so you can fine-tune or change it later?

**Quant tools:** what do you need to shrink a huge model down to run on smaller rigs? (llama.cpp, quant scripts, calibration data, etc.)

**How are you storing it?** Hashes, mirrors, torrents, offline drives?

Curious what people are actually doing, not looking for a politics fight.

(Currently using a Studio M2 192gb)

document fetched 2026-11-14 00:47 · canonical https://reddit.com/r/LocalLLaMA/comments/1xuthme/if_open_weights_dry_up_what_are_you_archiving