ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

update deepspeed install and run guide in README (#9441)

Browse source
This commit is contained in:
Heyang Sun 2023-11-17 09:11:39 +08:00 committed by GitHub
parent 30abd304a7
commit 921b263d6a
1 changed files with 3 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
python/llm/example/CPU/Deepspeed-AutoTP/README.md
View file

@ -8,6 +8,8 @@ Install necessary packages (here Python 3.9 is our test environment):
bash install.sh bash install.sh
``` ```
The first step in the script is to install oneCCL (wrapper for Intel MPI) to enable distributed communication between deepspeed instances, which can be skipped if Inte MPI/oneCCL/oneAPI has already been prepared on your machine. Please refer to [oneCCL](https://github.com/oneapi-src/oneCCL) if any related issue when install or import.
#### 2. Initialize Deepspeed Distributed Context #### 2. Initialize Deepspeed Distributed Context
Like shown in example code `deepspeed_autotp.py`, you can construct parallel model with Python API: Like shown in example code `deepspeed_autotp.py`, you can construct parallel model with Python API:
@ -57,7 +59,7 @@ If you want to run your own application, there are **necessary configurations in
```bash ```bash
# run.sh # run.sh
source bigdl-nano-init source bigdl-llm-init
unset OMP_NUM_THREADS # deepspeed will set it for each instance automatically unset OMP_NUM_THREADS # deepspeed will set it for each instance automatically
source /opt/intel/oneccl/env/setvars.sh source /opt/intel/oneccl/env/setvars.sh
...... ......