Promethevs: Full DeepSeek-v4.1-Flash-1017, locally, on 12GB of VRAM

Automated excerpt

Every word is first computed on a 2-bit copy of the model's experts. Engine Apache-2.0; report, figures and logs CC BY 4.0. Measured on the build machine, 160 new tokens per prompt, two passes per run.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.