LectureRechercheRadarCadre d'investissement
Connexion / Inscription
← Retour à la lecture
ActualitésSujet · DeepSeek推理框架Matériel1 min de lectureRédaction DisconfirmAI

DeepSeek releases open-source inference framework Deepinfer to cut self-hosting costs

DeepSeek launched Deepinfer, an open-source tool designed to reduce compute and energy costs for enterprises self-hosting AI models, currently limited to specific high-end hardware.

2026-10-21 22:00 GMT+8

Chinese AI company DeepSeek released an open-source inference framework named Deepinfer on GitHub on October 21, 2026. According to a report by German tech outlet t3n, the tool aims to help enterprises that deploy and run large language models themselves to lower their computing and energy costs, making AI usage more affordable.

Deepinfer optimizes the inference process through three main technical approaches: distributing computational tasks more evenly across available processor cores; utilizing 'Speculative Decoding' to speed up response times; and better structuring model outputs before final delivery. Currently, the framework only supports DeepSeek's own V4.1-Flash model, with plans to expand to other proprietary models in the future.

It is important to note that Deepinfer is not intended for general consumers or web-based users, but rather for enterprise clients with high-performance server clusters. It requires underlying hardware equipped with Nvidia GPUs based on the SM100 architecture or Huawei's Ascend 950 NPUs. These chips are typically offered as part of complete server systems to enterprises and are expensive; for instance, a single Huawei Ascend 950 chip costs over $37,000. Consequently, this tool primarily serves large tech companies or cloud service providers with significant self-hosting capabilities.

Sources

Connexion / Inscription
LectureRechercheRadarCadre d'investissement
t3n.de
↗
Sujet

DeepSeek推理框架

中国AI公司DeepSeek于2026年10月20日正式开源其内部生产级推理框架DeepInfer及配套组件deepseek-recipe,旨在降低大规模模型部署的算力与能源成本。该框架支持英伟达Blackwell(SM100)和华为昇腾9…

4 étapes5 piècesMis à jour 2026-11-07 19:16

Comment on en est arrivé là

  1. 2026-10-20DeepSeek宣布开源内部推理框架DeepInfer及编解码组件deepseek-recipe,支持英伟达与华为平台。ithome.com ↗
  2. 2026-10-20SiliconANGLE报道DeepSeek开源DeepInfer,指出部分核心调度代码暂未公开,并提及美国国土安全部此前要求联邦机构移除DeepSeek产品。siliconangle.com ↗
  3. 2026-10-21Latent Space发布AINews深度解析,披露DeepSeek声称的单集群日吞吐155万亿Token及每GPU小时38美元营收数据,引发关于利润率与回本周期的争议。latent.space ↗

Ce qui reste non résolu

  • DeepInfer未公开的KV缓存管理和负载均衡代码何时能完全解耦并开源?
  • vLLM等现有开源引擎能否通过配置优化达到DeepInfer宣称的效率水平?
  • DeepSeek披露的经济数据(如38美元/GPU小时营收)在不同工作负载和硬件成本下的实际利润率是多少?

Également dans ce fil

  • Latent Space AINews Analyzes DeepSeek's Open-Sourced Inference Engine DeepInfer: Details on 155T Daily Tokens and $38/GPU-Hour Revenue2026-10-23
  • DeepSeek Open-Sources DeepInfer: Compatible with Nvidia and Huawei Chips, Claims $38/Hour Revenue Per GPU2026-10-21
Lire le sujet complet →
Continuer la recherche →