2026-11-08 · America/Los_Angeles · 社区动态 · #18

For people using open-weight models in production, why did you choose them?

I've been experimenting with a fine-tuned Qwen vision model, and it got me thinking about why people choose open-weight models instead of just using hosted APIs. Is it mainly cost, privacy, being able to run models on your own infrastructure, avoiding dependence on one provider, or having the freedom to customize them? I'm especially curious about people building actual products or internal tools. Are you mostly serving existing models as-is, or have you needed to fine-tune them for your use cases? And has managing the models yourself been worth the effort, or are existing hosting services…

热度 44.3 / 100；排名与评分保留该期记录。