🧰 DeepInfer

The items that name it 6

• [AINews] OpenAI’s Patch the Planet program is 1 of 2 parties to discover PHP2Shell vulnerabilities in core PHP affecting 70% of the web, using GPT-5.6 Sol; coordinated response over 2 months
  articleas frameworks and tools2026-10-22

  DeepSeek's open-source production inference framework combining speculative decoding, dynamic prefill sizing, decode balancing by request count and KV length, and CPU/GPU scheduling overlap.

• SemiAnalysis (@SemiAnalysis_) on X
  threadas models2026-10-21
• [AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
  articleas frameworks and tools2026-10-21

  DeepSeek's open-source (Apache 2.0) production inference engine behind V4.1 Flash; supports distributed inference, speculative decoding, SWA Bounded Replay, structured output, EP load balancing, async scheduling on NVIDIA SM100 and Huawei Ascend. KV-cache management, PD disaggregation and load balan

• [AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
  articleas frameworks and tools2026-10-21

  DeepSeek's open-source inference engine behind V4.1 Flash, built on vLLM. Supports distributed inference, speculative decoding, SWA Bounded Replay, structured output, batch-invariant inference and deterministic sampling on NVIDIA SM100 and Huawei Ascend 950. KV-cache management, prefill/decode disag

• Latest open artifacts (#25): DeepSeek V4.1 Flash 1017, American-made models, and new regulatory controls in the US & China
  articleas frameworks and tools2026-10-20

  DeepSeek's internal inference stack, released publicly. Likely far more efficient than vLLM or SGLang for DeepSeek models; estimates put DeepSeek's gross margin on tokens served with it at up to 90%.

• DeepInfer: High-performance inference framework for DeepSeek models
  repoas frameworks and tools2026-10-20

  Open-source, high-performance inference framework that powers DeepSeek's large-scale production services. This release provides offline inference only, under the Apache License 2.0.