{"name":"deepseek-ultrafast","repo":"basetenlabs/deepseek-ultrafast","url":"https://midorreal.com/project/basetenlabs-deepseek-ultrafast","repository":"https://github.com/basetenlabs/deepseek-ultrafast","description":"","facts":{"language":{"value":"Rust","source":"https://github.com/basetenlabs/deepseek-ultrafast","checked":"21 Nov 2026"},"releases":{"value":"0","source":"https://github.com/basetenlabs/deepseek-ultrafast/releases","checked":"19 Nov 2026"},"contributors":{"value":"1","source":"https://github.com/basetenlabs/deepseek-ultrafast/graphs/contributors","checked":"19 Nov 2026"},"archived":{"value":"no","source":"https://github.com/basetenlabs/deepseek-ultrafast","checked":"21 Nov 2026"}},"stars":487.0,"writeup":{"what_it_is":"DeepSeek Ultrafast is an inference engine for DeepSeek-V4.1-Flash on sm_100a and sm_103a GPUs. It supports a 1M context length and shares KV cache across multiple servers.","audience":"Developers running DeepSeek-V4.1-Flash on NVIDIA GB300, B300 or B200 GPUs","claims":[{"kind":"specific","claim":"Decodes at 1,740 tokens/s on 4 x GB300 (TP4) for a coding prompt","excerpt":"**1,740 tokens/s** | **2,060 tokens/s**","status":"not_checked"},{"kind":"specific","claim":"Prefill of 1,000,000 tokens takes 13.2 s on 4 x GB300","excerpt":"1,000,000 tokens in 13.2 s","status":"not_checked"},{"kind":"specific","claim":"Supports 1M context length","excerpt":"It supports 1M context length and multi-server KV cache sharing.","status":"not_checked"},{"kind":"specific","claim":"Averages 5.6 tokens per decode step","excerpt":"**Acceptance length** | **5.6 tokens/step**","status":"not_checked"},{"kind":"specific","claim":"Server fetches KV cache from another server over RDMA instead of recomputing the prompt","excerpt":"that server fetches the cache directly from the server that holds it","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-14"},"why_now":null,"owner_supplied":null,"score":{"verdict":"too_early","hype":94,"reality":48,"gap":46,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T07:00:00+00:00","verdict_source":"rule"},"early_read":{"estimate":true,"lean":"too_close","gap_low":22,"gap_high":64,"rule":"early-2.0"},"changes":[{"date":"2026-10-27T06:10:13+00:00","text":"Repository created","source":"https://github.com/basetenlabs/deepseek-ultrafast"}],"cite":"deepseek-ultrafast is a GitHub project at https://github.com/basetenlabs/deepseek-ultrafast written mainly in Rust.","attribution":"Data from Mid or Real, https://midorreal.com/project/basetenlabs-deepseek-ultrafast, checked 22 November 2026. Attribution with a link is required.","still_real":null}