{"name":"DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks","repo":"MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks","url":"https://midorreal.com/project/miaai-lab-deepseek-v4-1-flash-1017-2x-dgx-sparks","repository":"https://github.com/MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks","description":"DeepSeek-V4.1-Flash-1017 EXL3 on 2x DGX Sparks with TensorFold","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks","checked":"22 Nov 2026"},"language":{"value":"Python","source":"https://github.com/MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks","checked":"22 Nov 2026"},"homepage":{"value":"https://mia-ai.net","source":"https://mia-ai.net","checked":"22 Nov 2026"},"releases":{"value":"0","source":"https://github.com/MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks/releases","checked":"21 Nov 2026"},"contributors":{"value":"19","source":"https://github.com/MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks","checked":"22 Nov 2026"}},"stars":499.0,"writeup":{"what_it_is":"A recipe for serving the DeepSeek-V4.1-Flash-1017 model in EXL3 2.9 bit quantization from two NVIDIA DGX Spark systems, linked by ConnectX-7 ports, using the TensorFold engine. It exposes an OpenAI-compatible API on port 8888 with a 1,048,576-token window, image input, tool calling, and speculative decoding.","audience":"Users with two DGX Spark or GB10 systems who want to self-host DeepSeek-V4.1-Flash","claims":[{"kind":"specific","claim":"Serves the model through an OpenAI-compatible API with a 1,048,576-token window","excerpt":"through an OpenAI-compatible API, with the full **1,048,576-token window**","status":"not_checked"},{"kind":"specific","claim":"Runs on two DGX Sparks with 128 GB each, one rank per Spark","excerpt":"Serve **DeepSeek-V4.1-Flash-1017** from two NVIDIA DGX Sparks (GB10, 128 GB each","status":"not_checked"},{"kind":"specific","claim":"Reports measured decode speed of 80.7 tok/s for one stream on prose","excerpt":"80.7 (TTFT 133 ms)","status":"not_checked"},{"kind":"specific","claim":"Reports measured prefill speed of about 2,100 tok/s","excerpt":"| tok/s | 2,105.9 | 2,152.0 |","status":"not_checked"},{"kind":"specific","claim":"Uses speculative decoding from the checkpoint's own draft blocks","excerpt":"**DSpark** speculative decoding from the checkpoint's own draft blocks","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-22"},"why_now":null,"owner_supplied":null,"score":{"verdict":"real","hype":59,"reality":64,"gap":-4,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-10-18T18:10:59+00:00","text":"Repository created","source":"https://github.com/MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks"}],"cite":"DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks is a GitHub project at https://github.com/MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks written mainly in Python, described there as \"DeepSeek-V4.1-Flash-1017 EXL3 on 2x DGX Sparks with TensorFold\".","attribution":"Data from Mid or Real, https://midorreal.com/project/miaai-lab-deepseek-v4-1-flash-1017-2x-dgx-sparks, checked 22 November 2026. Attribution with a link is required.","still_real":null}