Advanced Technologies: Beyond Prompting - Choosig a model

Model size:

  • 1B parameters:--classification
    • Pattern matching and basic knowledge of the world.
  • 10B parameters:-- chatbot
    • Greater world knowledge. Can follow basic instructions, but just knows more esoteric facts about the world.
    • especially if you were to fine-tune it to become better at the types of specific instructions you wanted to follow
  • 100B+ parameters: -- brainstorming partner
    • Rich world knowledge. Complex reasoning.
    • know a lot of things about physics, philosophy, history, science, and so on,

Closed or open source?

  • Closed-source models
    • Cloud programming interface
    • Easy to use in applications
    • More large/powerful models
    • Relatively inexpensive
    • Some risk of vendor lock-in
  • Open-source models
    • Full control over the model
    • Can run on your own device(on perm, PC, etc.)
    • Full control over data privacy/access
相关推荐
AiNightVision1 分钟前
无人机夜视模块如何选择?——从参数指标到系统集成的完整选型指南
人工智能·算法·计算机视觉·车载系统·无人机·视频·智能硬件
Java后端的Ai之路10 分钟前
【AIAgent开发】-Agent工具封装成CLI(纯干货!!!)
java·javascript·人工智能·agent·cli
Flynn Liang26 分钟前
Spring AI 上生产:限流熔断、密钥与成本治理、私有化/混合部署、降级与 Prompt 灰度架构全解
人工智能·spring·架构·高可用·springai·成本治理·生产架构
飞猫的边缘AI35 分钟前
边缘AI芯片的规则变了:从跑CNN到跑LLM和VLM
人工智能·语言模型·cnn·transformer·边缘ai
code2cat1 小时前
【随笔】连续批处理:长短请求怎样在迭代之间补位
java·人工智能·后端·大语言模型·vllm
牧羊人.3331 小时前
卷积神经网络之VGG
人工智能·深度学习·神经网络·算法·cnn
Flynn Liang1 小时前
Spring AI 2.0 深度解析:ChatClient、Advisor 链与 MCP——Java 工程师的 AI 落地最优解
java·人工智能·spring·大模型·架构设计·spring ai
Martina_03211 小时前
AI生成3D模型一合并网格,材质就串色?用6步检查材质槽、UV与图集重映射
图像处理·人工智能·3d·材质·uv·游戏策划·1024程序员节
蓝鲨硬科技1 小时前
Token之后,物理AI需要什么样的度量衡?
大数据·人工智能
人工智能AI技术1 小时前
Agent架构新思路:多Agent正在过时,单循环Main Loop才是新方向
人工智能