# 大模型

[大语言模型部署可行性分析](https://hackmd.io/@-iruZ-_WSi6OcadE3FUrdQ/r1VpPM4v3)

## 大模型应用

### Prompt

## 推理

[vLLM](https://vllm.readthedocs.io/en/latest/)

[vllm-project/vllm](https://github.com/vllm-project/vllm)

## LLM

## 极客时间

[AIGC知识库](https://longalong.feishu.cn/wiki/wikcngNXeTD1ZnOly0X19lPLfoh)

## 框架

### 数据并行

[大模型技术之数据并行](https://www.bilibili.com/video/BV1WP41197DR/?spm_id_from=333.999.0.0&vd_source=83f3b8f8e4d0dd8ef1cfa1a66c9802f7)

### 模型并行

[大模型技术之模型并行](https://www.bilibili.com/video/BV14u41147MP/?spm_id_from=trigger_reload&vd_source=83f3b8f8e4d0dd8ef1cfa1a66c9802f7)

### Pipeline并行

[大模型技术之Pipeline并行](https://www.bilibili.com/video/BV14m4y1t7YG/?spm_id_from=333.788.recommend_more_video.1&vd_source=83f3b8f8e4d0dd8ef1cfa1a66c9802f7)

### ZeRO

[大模型技术之优化器并行(ZeRO)](https://www.bilibili.com/video/BV1od4y1Z7kf/?spm_id_from=333.337.search-card.all.click&vd_source=83f3b8f8e4d0dd8ef1cfa1a66c9802f7)

[训练一个130亿参数的模型要用几个GPU？微软：一个就够](https://zhuanlan.zhihu.com/p/347068396)

[分布式训练 Parameter sharding 之 ZeRO](https://www.cnblogs.com/rossiXYZ/p/15785669.html)

### 重计算

[大模型技术之重计算](https://www.bilibili.com/video/BV1xg4y1c7Cw/?spm_id_from=333.999.0.0&vd_source=83f3b8f8e4d0dd8ef1cfa1a66c9802f7)

### 指令微调

[【大模型技术】指令微调](https://www.bilibili.com/video/BV1Do4y177HW/?spm_id_from=333.788.recommend_more_video.-1&vd_source=83f3b8f8e4d0dd8ef1cfa1a66c9802f7)

## MindSpore

[昇思MindSpore](https://www.mindspore.cn/)