[Mthreads] support llama 13B train #9666

shang-mt · 2024-12-20T02:32:31Z

PR types

PR changes

Description

paddle-bot · 2024-12-20T02:32:36Z

Thanks for your contribution!

CLAassistant · 2024-12-20T02:32:37Z

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

ZHUI · 2024-12-20T02:59:35Z

llm/mthreads/llama/README.md

+```
+
+### （2）训练：
+1. 多机多卡推理


咱们不是推理，是训练吧

已修改。

ZHUI · 2024-12-20T03:00:23Z

llm/mthreads/llama/README.md

+
+执行如下命令进行推理：
+```bash
+bash run_dist.sh 10.10.10.123 # 假设master ip 为10.10.10.123，在不同节点上执行此命令


到哪个目录运行run_dist.sh 脚本，需要说明一下，用户可能找不到。

已添加进入目录命令

ZHUI · 2024-12-20T03:01:43Z

llm/mthreads/llama/README.md

+3. 安装 paddle
+```
+# paddlepaddle『飞桨』深度学习框架，提供运算基础能力
+git clone git@github.com:PaddlePaddle/Paddle.git -b release-musa/2.6


依赖的paddle代码都合入到了这个官方分支了吗？这个分支可以直接跑?

是的，这个代码已经都合入完成了。

ZHUI · 2024-12-20T03:02:36Z

llm/mthreads/llama/README.md

+4. 克隆 PaddleNLP 仓库代码，并安装依赖
+```
+# PaddleNLP是基于paddlepaddle『飞桨』的自然语言处理和大语言模型(LLM)开发库，存放了基于『飞桨』框架实现的各种大模型，llama2-13B模型也包含其中。为了便于您更好地使用PaddleNLP，您需要clone整个仓库。
+git clone git@github.com:shang-mt/PaddleNLP.git -b mthreads-llama-13B


这个分支可以提一个pr到paddlenlp，不合入，留一下作为记录吧。

#9193
这个Pr之前有提交，可以直接跑。

ZHUI · 2024-12-20T03:04:16Z

llm/mthreads/llama/README.md

@@ -0,0 +1,77 @@
+## 🚣‍♂️ 使用 PaddleNLP 在 MTT S4000 下跑通 llama2-13b 模型 🚣


Suggested change

## 🚣‍♂️ 使用 PaddleNLP 在 MTT S4000 下跑通 llama2-13b 模型 🚣

## 🚣‍♂️ 使用 PaddleNLP 在 MTT S4000 下跑通 llama2-13b 模型预训练 🚣

咱们应该支持了预训练，对吧

codecov · 2024-12-20T03:06:03Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 52.79%. Comparing base (19fcdcb) to head (a336de3).
Report is 268 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #9666      +/-   ##
===========================================
+ Coverage    52.56%   52.79%   +0.23%     
===========================================
  Files          721      718       -3     
  Lines       116012   112241    -3771     
===========================================
- Hits         60976    59262    -1714     
+ Misses       55036    52979    -2057

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

ZHUI

LGTM

paddle-bot bot added the contributor label Dec 20, 2024

paddle-bot bot assigned gongel Dec 20, 2024

shang-mt force-pushed the develop branch from 21eca11 to e0339de Compare December 20, 2024 02:37

ZHUI reviewed Dec 20, 2024

View reviewed changes

[Mthreads] support llama 13B train

a336de3

shang-mt force-pushed the develop branch from e0339de to a336de3 Compare December 20, 2024 03:46

ZHUI approved these changes Dec 20, 2024

View reviewed changes

ZHUI merged commit 297dbce into PaddlePaddle:develop Dec 20, 2024
9 of 12 checks passed

blacksheep-Aristotle pushed a commit to blacksheep-Aristotle/PaddleNLP that referenced this pull request Dec 23, 2024

[Doc] [Mthreads] Support llama 13B pre-train (PaddlePaddle#9666)

fdd25bc

		@@ -0,0 +1,77 @@
		## 🚣‍♂️ 使用 PaddleNLP 在 MTT S4000 下跑通 llama2-13b 模型 🚣

[Mthreads] support llama 13B train #9666

[Mthreads] support llama 13B train #9666

Uh oh!

Conversation

shang-mt commented Dec 20, 2024

PR types

PR changes

Description

Uh oh!

paddle-bot bot commented Dec 20, 2024

Uh oh!

CLAassistant commented Dec 20, 2024

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

shang-mt Dec 20, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

codecov bot commented Dec 20, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

ZHUI left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

shang-mt Dec 20, 2024 •

edited

Loading

codecov bot commented Dec 20, 2024 •

edited

Loading