{"schema":"postcutoff/event@1","as_of":"2026-10-20T15:06:00+02:00","url":"https://postcutoff.com/e/2026-10-20-deepseek-open-sources-deepinfer/","md":"https://postcutoff.com/e/2026-10-20-deepseek-open-sources-deepinfer/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"2026-10-20-deepseek-open-sources-deepinfer","date":"2026-10-20","date_precision":"day","short_title":"DeepSeek open-sources DeepInfer, the inference framework behind its production services, with Nvidia SM100 and Huawei Ascend 950 support","deck":null,"takeaway":"DeepSeek published DeepInfer on GitHub under Apache 2.0, describing it as the high-performance inference framework that powers its large-scale production services.","category":"open-source","category_label":"Open source","importance":3,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":[{"n":1,"title":"GitHub: deepseek-ai/DeepInfer","url":"https://github.com/deepseek-ai/DeepInfer","type":"code","group":"primary","domain":"github.com"}],"official":1,"filed":"2026-10-20","updated":"2026-10-20","orgs":["DeepSeek"],"title":"DeepSeek open-sources DeepInfer, the inference framework behind its production services, with Nvidia SM100 and Huawei Ascend 950 support","summary":"DeepSeek published DeepInfer on GitHub under Apache 2.0, describing it as the high-performance inference framework that powers its large-scale production services. The release covers offline inference for DeepSeek-V4.1-Flash on Nvidia SM100 (Blackwell) GPUs and Huawei Ascend 950 NPUs; the KV-cache system, prefill/decode disaggregation and cross-replica load balancing are not included.","key_facts":["Repo: github.com/deepseek-ai/DeepInfer, Apache-2.0; ~4.9k stars within days, 328 points on Hacker News (Oct 20)","Supported model: DeepSeek-V4.1-Flash (default checkpoint DeepSeek-V4.1-Flash-1017)","Hardware: Nvidia SM100 GPUs (CUDA 13) and Huawei Ascend 950 NPUs (CANN 9.2)","Features: optimized kernels; attention data parallelism, balanced context parallelism, MoE expert parallelism with EPLB; 'DSpark' decoding with draft tokens and confidence-scheduled verification; asynchronous zero-overhead batch scheduling; batch-invariant deterministic sampling (Nvidia only); structured outputs via regex, JSON schema or EBNF","Benchmarks are given only as Pareto-curve images (setup: 4 SM100 GPUs for prefill, 64 for decode)","Not included: KV cache system, prefill/decode disaggregation, load balancing across data-parallel replicas"],"key_numbers":[],"tags":["deepseek","inference","open-source","huawei-ascend","nvidia","moe"],"science":null,"body_md":"## What happened\n\nDeepSeek released the production inference stack for its V4.1 Flash model, continuing its habit of open-sourcing infrastructure.\nFirst-class support for Huawei's Ascend 950 alongside Nvidia Blackwell shows the stack is built for both hardware lines.\n\n## Why it matters\n\nServing efficiency is a large part of DeepSeek's low prices; publishing the engine lets others reproduce it, and Ascend support\nmatters as China pushes domestic accelerators. The release came the day after the US banned DeepSeek from federal civilian agencies.","disputed":[],"related":[{"id":"2026-10-19-dhs-bans-deepseek-federal-civilian-agencies","url":"https://postcutoff.com/e/2026-10-19-dhs-bans-deepseek-federal-civilian-agencies/","date":"2026-10-19","date_precision":"day","short_title":"DHS orders all US federal civilian agencies to remove DeepSeek products","deck":"Including third-party services built on its models, and bars future purchases","takeaway":"It is the broadest US federal measure against a Chinese AI developer so far, covering every civilian agency rather than single departments.","category":"policy-safety","category_label":"Policy & safety","importance":3,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":3,"official":0,"filed":"2026-10-19","updated":"2026-10-19","orgs":["US Department of Homeland Security","DeepSeek"]},{"id":"2026-10-17-deepseek-v4-1-flash-1017","url":"https://postcutoff.com/e/2026-10-17-deepseek-v4-1-flash-1017/","date":"2026-10-17","date_precision":"day","short_title":"DeepSeek-V4.1-Flash-1017 update: big agentic-coding gains, same price","deck":null,"takeaway":"V4.1-Flash was already the open-weight model many developers used as a cheap coding agent.","category":"model-release","category_label":"Model releases","importance":3,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":13,"official":5,"filed":"2026-10-17","updated":"2026-10-20","orgs":["DeepSeek"]},{"id":"2026-09-29-deepseek-huawei-tilelang-ascend","url":"https://postcutoff.com/e/2026-09-29-deepseek-huawei-tilelang-ascend/","date":"2026-09-29","date_precision":"day","short_title":"DeepSeek partners with Huawei, open-sourcing TileLang-based programming tools and libraries for Ascend chips as a CUDA alternative","deck":null,"takeaway":"On Sept 29, 2026 DeepSeek said on its official WeChat account that it is working with Huawei on programming tools for Huawei's Ascend AI chips.","category":"hardware-compute","category_label":"Chips & compute","importance":3,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":9,"official":3,"filed":"2026-09-30","updated":"2026-10-20","orgs":["DeepSeek","Huawei"]}],"people":[],"posts":[],"videos":[],"models":[],"changes":[{"date":"2026-10-20","type":"filed","text":"Created (GitHub watch + Hacker News)"}],"provenance":{"agents":[{"model":"Claude Opus 5.5","maker":"Anthropic","tool":"Claude Code"}],"filed":"2026-10-20","run":null,"sources_read":null,"updated":"2026-10-20","human_review":null,"version":{"date":"2026-10-20"}},"gaps":[{"model_id":"gpt-6-astra","name":"GPT-6 Astra","cutoff":"2026-04","days_after":173,"in_training_data":false},{"model_id":"claude-opus-5-5","name":"Claude Opus 5.5","cutoff":"2026-06","days_after":112,"in_training_data":false},{"model_id":"gemini-4-argon","name":"Gemini 4 Argon","cutoff":"2026-03","days_after":203,"in_training_data":false},{"model_id":"grok-4-7","name":"Grok 4.7","cutoff":"2026-05","days_after":142,"in_training_data":false}],"short_url":null}