Replies: 27 comments 28 replies
-
发展方向:由专用模型到通用模型 第一步模型选型。针对于应用场景,比较多种大模型相关维度的能力,进行模型评测。经过模型评测初步选型之后,可选定意向大模型。 第二步评估业务场景复杂度。若不复杂,可直接把模型拿过来应用。则继续如果业务场景非常复杂,通常来讲直接开源模型无法满足需求,需要微调、prompt工程等进一步构建。 第三步,判断全参数微调或部分微调等,并计算所需算力,应提前规划,避免微调失败。如果算力足够,可进行全参数微调。如果算力资源比较受限,只能进行部分参数微调,类似把大部分的参数固定住,只调一小部分参数。 第四步,构建基于大模型的智能体需要考虑模型与环境的交互。如果需要调用外部API或与已有业务数据库交互,就需要构建智能体。如果不需要与环境交互,就可以直接将微调好的模型在业务场景中试用。 第五步,模型评测,并评估是否上线应用或继续迭代。 第六步,模型部署。关于软件系统相关性能、安全、功能等方面内容。如考虑如何以更少的资源部署模型,或者如何提升整个应用的吞吐量。 |
Beta Was this translation helpful? Give feedback.
-
https://github.com/LordHidy/InternLM_learning/blob/main/Notes/Lesson_1.md |
Beta Was this translation helpful? Give feedback.
-
https://zqm6a0l4f0l.feishu.cn/wiki/Pv7VwKByIim5OfkPaLScc172ndg?from=from_copylink |
Beta Was this translation helpful? Give feedback.
-
https://blog.csdn.net/ynx005/article/details/135484860?spm=1001.2014.3001.5502 |
Beta Was this translation helpful? Give feedback.
-
https://blog.csdn.net/lalala12ll/article/details/135487885?spm=1001.2014.3001.5501 |
Beta Was this translation helpful? Give feedback.
-
https://p065drvsp5m.feishu.cn/docx/SnXldHZkMoeNQLx2F7qcgsRzn1c?from=from_copylink |
Beta Was this translation helpful? Give feedback.
-
https://blog.csdn.net/jinghangyz/article/details/135488141?spm=1001.2014.3001.5502 |
Beta Was this translation helpful? Give feedback.
-
https://github.com/zengya55/InterLM_Learning/blob/main/lesson1/InterLM%20course%201.md |
Beta Was this translation helpful? Give feedback.
-
https://blog.csdn.net/weixin_52524888/article/details/135491112 |
Beta Was this translation helpful? Give feedback.
-
https://blog.csdn.net/zhh763984017/article/details/135491600 |
Beta Was this translation helpful? Give feedback.
-
https://blog.csdn.net/weixin_48396757/article/details/135494823?spm=1001.2014.3001.5501 |
Beta Was this translation helpful? Give feedback.
-
https://github.com/ykuang/internlm-tutorial/blob/main/notes/lesson1.md |
Beta Was this translation helpful? Give feedback.
-
书生·浦语大模型系列 轻量级:InternLM-7B
数据:书生·万卷 文本数据:50亿个文档,数据量超1TB 高可扩展:支持从8卡到千卡训练,千卡加速效率达92%。 增量续训 全链路开源开放体系|微调 OpenCompass评测体系 全链路开源开放体系|部署 大语言模型特点 低存储设备:消费级显卡、移动端等如何部署? 如何加速token的生成速度 提升系统整体吞吐量 全链路开源开放体系|智能体 轻量级智能体Lagent 多模态智能体工具箱AgentLego 丰富的工具集合,尤其是提供了大量视觉、多模态相关领域的前沿算法功能 |
Beta Was this translation helpful? Give feedback.
-
第一节笔记 1609 |
Beta Was this translation helpful? Give feedback.
-
https://cewmxvz60v.feishu.cn/wiki/K5tDwibmwiAsWckr9cpcZAipnwc |
Beta Was this translation helpful? Give feedback.
-
https://blog.csdn.net/2301_80135859/article/details/135540106 |
Beta Was this translation helpful? Give feedback.
-
第一课笔记_Wade |
Beta Was this translation helpful? Give feedback.
-
第一节课笔记 |
Beta Was this translation helpful? Give feedback.
-
请上传第1节课的笔记在此处~
Beta Was this translation helpful? Give feedback.
All reactions