NVIDIA's FP4 build of Qwen3. 8-Flash-Next is 96 GiB on disk. The BF16 original is about 355 GB.
Source: [Dev.to](https://dev.to/alanwest/how-to-run-qwen38-flash-next-fp4-locally-it-needs-blackwell-and-55-env-vars-3oii)
1 points, 0 comments on Hacker News
1 points, 0 comments on Hacker News
1 points, 0 comments on Hacker News
Most RAG tutorials end when the chatbot answers a question about a PDF. I wanted to find out what comes after that: how do you know it works, who is allowed to see what, what happens when a document tries to give it orders, and how do you ship it without paying for anything? This post walks thr...
For a long time my release process looked like this: build a feature, test it well on staging, and then be told it should go live after business hours. So I'd stay up, merge the pull request at the right time, watch the deploy, and be ready to roll it back. That's a boring job for a person.
Along with 'Artificial,' 'Musk' and 'The Social Reckoning,' the storytellers are finally getting back at the engineers.