DeepSeek-V4.1-Flash 技术报告解读：CED 与 CSA2 如何将 KV 缓存压缩至每 token 890 字节

WorkBuddy是面向国内用户的通用Agent产品，支持Windows和Mac，提供免费版和58元/月个人专业版，企业版已推出。内置代码开发、日常办公、设计创意三种场景模式及100多个行业领域AI专家。模型集成腾讯混元、DeepSeek（推荐V4 Pro）、GLM、Kimi等国产大模型，也支持接入兼容OpenAI协议的外部API。拥有Skills市场和MCP连接器生态，可打通QQ邮箱、腾讯会议、腾讯文档等服务。教程通过公众号周报生成和功能网页开发两个案例演示实际用法。

I added a WebMCP benchmark to DeepDeck, my MIT-licensed macOS desktop project built on DeepSeek...

I packaged DeepSeek Harness (dsh) into a double-clickable macOS installer for Apple Silicon. It worked on my machine — then died instantly on a clean Mac. Here's the otool -L investigation that found 24 absolute library paths baked into th…

A week ago I published a fairly absurd number: 612.9 million tokens for $28.35. That was DeepSeek...

OpenArch 用 PyTorch 从零手写实现现代开源 LLM 架构，每个架构一个可读文件，覆盖 GPT-2 XL、Llama 2/3、DeepSeek R1、Gemma 3、Llama 4 Maverick、Qwen 3、Kimi K2、GLM 4.5、GPT-OSS 等，并标注参数量、归一化、位置编码、注意力与 MoE 选择。

A deep mathematical and PyTorch breakdown of Multi-Head Latent Attention (MLA), matrix absorption, and decoupled RoPE.

NVIDIA 在 JAX 中用 Transformer Engine 优化 Dropless MoE 训练，在 NVIDIA GB200 上跑 DeepSeek-V3 时，未优化基线仅 103 TFLOPS/GPU、GPU 间通信占累计 kernel 时间 84%，经针对性 kernel 优化后升至 1，068 TFLOPS/GPU，提升 10.4 倍。

LLM Pricing Digest: DeepSeek Quietly Drops Two New Flash Models Weekly roundup from LLM...

In this article, I will explain how you can set DeepSeek as the default model for your Codex coding...

DeepSeek 开源了一些新的代码仓库，方便更容易地部署 V4.1 Flash 以及后续的开源模型： https://github.com/deepseek-ai/deepseek-recipe https://github.com/deepseek-ai/DeepSelect https://github.com/deepseek-ai/DeepJIT