shibing624 / MedicalGPT

MedicalGPT: Training Your Own Medical GPT Model with ChatGPT Training Pipeline. 训练医疗大模型,实现了包括增量预训练(PT)、有监督微调(SFT)、RLHF、DPO、ORPO。
Apache License 2.0
2.94k stars 452 forks source link

Question about hardware requirements - 关于硬件要求的问题 #325

Closed elieobeid7 closed 5 months ago

elieobeid7 commented 5 months ago

This hardware requirement section isn't clear

  1. Does this table show RAM or hard disk space? Or is it VRAM?
  2. How much RAM and hard disk space is it recommended to use?
  3. Which graphics card to use?
  4. Is there a small model we can download on PC to develop locally before we upload it to the server?

这个硬件要求部分不清楚

  1. 这个表格显示的是RAM(随机存取存储器)还是硬盘空间?
  2. 建议使用多少RAM和硬盘空间?
  3. 应该使用哪款显卡?
  4. 有没有一个小型模型,我们可以在PC上下载来本地开发,然后再上传到服务器?

This question has been translated into Chinese by ChatGPT as I don't speak Chinese 这个问题已经被ChatGPT翻译成中文,因为我不会说中文。

shibing624 commented 5 months ago
  1. VRAM
  2. 32G RAM cpu, and 200G disk
  3. A100 or V100
  4. you can use 6B or Qwen-1.8B: https://huggingface.co/Qwen/Qwen-1_8B
elieobeid7 commented 5 months ago

Ok thank you.

elieobeid7 commented 4 months ago

@shibing624 trying to run it on AWS EC2 p4d.24xlarge, it complains about Ampere GPU when doing sh run_pt.sh

image

Only A100 processors are based on Ampere architecture. AWS have only p4d.24xlarge instances with A100

Does it run on p4d.24xlarge?

shibing624 commented 4 months ago

use fp16