←搜索结果

搜索“RAG”

按匹配度 · 共 384 条结果 清除搜索

星期一

  1. Claude Platform：开发者版本说明

    Anthropic 开发者控制台新增提示词生成器

    Anthropic 在开发者控制台推出提示词生成器，用户描述任务后 Claude 会运用角色设定、链式推理等技巧生成可编辑的提示词模板。该功能基于一个长提示词实现，并支持将模板变量用于评估，ZoomInfo 称其将提示词优化时间减少了 80%。

    est practices The prompt templates generated by this new feature make use of many of our prompt engineering best practices. One such practice is role setting, where Claude is encouraged to take on the characteristics of an expert at the chosen task. In our content moderation example, the role setting looks like this: Another practice is chain of thought reasoning, in which Claude is given time and space to collect its thoughts before answering. This allows for more thorough and well-reasoned res

    匹配字段正文

    #company#产品更新#Anthropic

星期四

  1. Ollama Blog精选

    Ollama 发布 Windows 预览版

    Ollama 在 Windows 上推出预览版，可在原生 Windows 环境中拉取、运行和创建大语言模型。该版本内置 GPU 加速，支持 NVIDIA GPU 以及 AVX、AVX2 等 CPU 指令集，无需配置或虚拟化；同时提供完整模型库（含 LLaVA 1.6 等视觉模型，图片可拖入 ollama run）和后台运行的 Ollama API（http://localhost:11434），并保持与其他平台一致的 OpenAI 兼容性。

    Ollama windows

    ull access to the model library The full Ollama model library is available to run on Windows, including vision models. When running vision models such as LLaVA 1.6, images can be dragged and dropped into ollama run to add them to a message. Always-on Ollama API Ollama’s API automatically runs in the background, serving on http://localhost:11434. Tools and applications can connect to it without any additional setup. For example, here’s how to invoke Ollama’s API using PowerShell: Ollama on Window

    匹配字段正文

    #company#产品更新#部署/工程#OpenAI

星期二

  1. vLLM 官方博客精选

    vLLM 发布基于 PagedAttention 的开源 LLM 推理与服务库

    vLLM 发布开源 LLM 推理与服务库，利用 PagedAttention 管理 KV cache，无需修改模型架构即可实现比 HuggingFace Transformers 最高 24x 的吞吐量。实验中其吞吐量比 TGI 最高提升 3.5x，内存浪费降至 4% 以下；LMSYS 部署后将服务所需 GPU 数量减少 50%。

    highly variable and unpredictable. As a result, efficiently managing the KV cache presents a significant challenge. We find that existing systems waste 60% – 80% of memory due to fragmentation and over-reservation To address this problem, we introduce PagedAttention, an attention algorithm inspired by the classic idea of virtual memory and paging in operating systems. Unlike the traditional attention algorithms, PagedAttention allows storing continuous keys and values in non-contiguous memory sp

    匹配字段正文

    #engineering#产品更新#部署/工程#开源生态

星期四

  1. Replit Blog精选

    Replit 发布 Ghostwriter AI 编程助手，Complete Code 进入封闭测试

    Replit 宣布推出 Ghostwriter，把 AI 能力接入几乎全部 IDE 功能，旗舰功能 Complete Code 是实时补全代码的 AI 结对程序员，目前处于封闭测试。官方称通过 FasterTransformer 与 Triton 推理、知识蒸馏和量化等优化，中位响应时间低于 400ms；Explain Code、Generate Code、Transform Code 则调用 OpenAI 等更大模型。Ghostwriter 后续将打包为基于 Cycles 的付费功能，最终计划像 Multiplayer 一样默认免费开放。

    Ghostwriter AI & Complete Code Beta — cover image

    e in order to generate entire programs or files. That kind of generations benefits from the performance of insanely large models (100B+ parameters). Transform Code Finally, we leverage exceptionally large models to provide prompt-driven refactor/rewrite experiences. The future With the advent of LLMs and generative models in general, we believe that software is entering a new epoch. In the near future, anyone with time and good ideas will be able to build amazing things. AI will guide you as you

    匹配字段正文

    #company#产品更新#编码#部署/工程