DeepSeek V4.1 Flash 1017 original weights running locally!
Paper Journey Through Japan - mlx-server 2xM3 Ultra A Day in the Forest - TensorFold 4xDGX Sparks DeepSeek Harness used: - cache hit rate: 99% - average decode speed on both side has been ~ 52 t/s - Max context used
46 models. 1 API key. 100% OpenAI-compatible.
Just dropped the full dulus-* model lineup on DULUS — from dulus-x-397b, dulus-deepseek-v4-pro, and dulus-kimi-k2.7-code to dulus-glm-5.3, dulus-llama-4-scout, and dulus-free tiers. Switch models in a split second inside the Dulus
Boop: a personal AI assistant running on DeepSeek Harness, living on your computer and responding to you on your phone.
After installing it, scan the code once and it can be used anywhere. It responds to you with maps, forms, and tables, all of which can be clicked directly. It has its own browser and command line, and you can continue working even after closing the app. Open source, the interface is written by @jxnlco's NSH. All-inclusive, generative AI, all in one place.
You have reached the end of the archive
All of deepseek