ok this one's nerdy but I love it. Josef Gattermayer wrote AMD kernels for EXL3, the…
This is a AI post classified by AGK Intelligence as Open & local models (a model release), kept by the AI Radar because it carries real work, not commentary.
ok this one's nerdy but I love it. Josef Gattermayer wrote AMD kernels for EXL3, the quant format that's pretty much been NVIDIA-only, and got them into Project Maya. DeepSeek V4.1 Flash at 2.25 bpw, 1M context, on a Strix Halo (Ryzen AI Max+ 395) plus two secondhand RX 7900 XTXs over OCuLink. 59.6 tok/s on Maya's own bench, and it read a 1,040,000-token prompt at 980 tok/s. best detail... Linux memory compaction was freezing both GPUs for up to 300 ms, one sysctl fixed it. there's a one-command Ansible install too
Posted by Mike Gannotti 🔳 (54.8k followers) 1 h ago · 4 likes · 155 views · view the original post on X. Kept by the AI Radar as Open & local models.
More AI work like this
- Another magic by DeepSeek v4.1 + Blender MCP + DeepSeek Harness running locally! — @ivanfioravanti
- 🥱 Tired of your #ai open-weight #LLM answering free speech questions like it's… — @michabbb
- You can now Run Uncensored DeepSeek-V4-Pro-Qwen3.5-9B locally on a 5 GB Laptop with zero… — @0x0SojalSec
- You can run locally Uncensored "DeepSeek-V4-Pro-Qwen3.5-9B" 5.6GB. — @0x0SojalSec
- Meet the Qwen3.8-27B Ultimate Heretic Uncensored Turbo Brilliance 50x AGI NEO GGUF. It's… — @HuggingModels
- Battle 2 of my AI tournament! ⚔️ — @niklaslenz_ai
- Comparing Ameriqwen against Alibaba Qwen — @slimer48484
- Meet Mirage, a new image-text-to-text model from Cactus-Compute. It handles object… — @HuggingModels
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 65.3k posts from 5k X accounts over the last 14 days, 10.6k tools, 19 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-07 22:45 UTC. Full method.