Dongxi 东锡 NLP· @dongxi_nlp · X·· 2 小时前AI 评分63
AI 导读
DeepSeek 以 Apache-2.0 协议开源了高性能大规模推理框架 DeepInfer，代码已发布在 GitHub。该框架包含优化后的 kernel、模型与采样器运行时，支持 Attention DP 与均衡 CP、MoE EP 加 EPLB 等并行策略，以及 DSpark 与置信度调度验证、SWA 有界重放、零开销批调度器、批不变推理、确定性采样和结构化输出。
正文
DeepSeek 开源 DeepInfer
Respect!
DeepSeek just open sourced its high-performance, large-scale inference framework DeepInfer in Apache-2.0 license. https://github.com/deepseek-ai/DeepInfer Supported features: - Optimized kernels, model and sampler runtime - Parallelism: Attention DP / balanced CP, MoE EP + EPLB - DSpark and confidence-scheduled verification - SWA bounded replay - Asynchronous scheduling (zero-overhead batch scheduler) - Batch-invariant inference - Deterministic sampling - Structured output在 X 查看被引用的帖子
来源：Dongxi 东锡 NLP · x.com