DeepSeek-V4.1-Flash EXL3 is finally alive across 4× DGX Sparks. 🔥
This is a dev post classified by Jev as AI dev tools (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
DeepSeek-V4.1-Flash EXL3 is finally alive across 4× DGX Sparks. 🔥 I built the quality-first 4.75 bpw SAGE quant. @cfontes and @Blackwellboy spent more than a week, 14 PRs, dozens of commits and 40+ bring-up iterations making it actually run. I don’t own one Spark, let alone four. This is what open-source collaboration looks like. 𝗧𝗛𝗘 𝗥𝗘𝗖𝗘𝗜𝗣𝗧𝗦 Single-stream decode: 30.2 tok/s prose 33.4 tok/s code 33.37 tok/s best warm sustained cell Aggregate at c=1 / 2 / 4 / 8: 27.8 / 43.9 / 57.7 / 61.5 tok/s KV budget: 3.97M tokens Context: full 1,048,576 tokens Every figure is a sustained 30-secon
Posted by Cruz (2.1k followers) 1 h ago · 4 likes · 338 views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: dsv4.1-flash-exl3-4.75bpw, deepseek-v4.1-flash-exl3-dgx-spark-recipe, vllm-exl3.
More dev work like this
- this is what 12gb of vram builds in 2026, absolute magic — @sudoingX
- Today, we’re shipping updates to @Rippling AI that make it easier to deploy data… — @stanine
- Agent workflows are easy to demo. Operating them is the hard part. — @DanKornas
- Run open models like Gemma 4 completely offline in the Antigravity SDK. — @antigravity
- Wow… you can now vibe code apps for the iPhone Duo with GPT-6 Sol and Opus 5.5! — @rileybrown
- OpenAI GPT‑6 Sol and Luna are on the APEX leaderboards. — @mercor
- Terminal-Bench meetup tonight *with* swag. Register if you haven’t already! Benchmarks,… — @alexgshaw
- Getting access to quality healthcare often depends on where you live and whether a… — @NewsFromGoogle
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 22.8k posts from 5k X accounts over the last 21 days, 2.6k tools, 12 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-23 18:47 UTC. Full method.