polish readme

pull/2484/head
jiaruifang 2023-01-16 15:17:27 +08:00
parent 37baea20cb
commit 315e1433ce
2 changed files with 20 additions and 9 deletions

View File

@ -39,9 +39,15 @@ If you want to test ZeRO1 and ZeRO2 in Colossal-AI, you need to ensure Colossal-
For simplicity, the input data is randonly generated here. For simplicity, the input data is randonly generated here.
## Training ## Training
We provide two solutions. One utilizes the hybrid parallel strategies of Gemini, DDP/ZeRO, and Tensor Parallelism. We provide two stable solutions.
The other one uses Pipeline Parallelism Only. One utilizes the Gemini to implement hybrid parallel strategies of Gemini, DDP/ZeRO, and Tensor Parallelism for a huggingface GPT model.
In the future, we are going merge them together and they can be used orthogonally to each other. The other one use [Titans](https://github.com/hpcaitech/Titans), a distributed executed model zoo maintained by ColossalAI,to implement the hybrid parallel strategies of TP + ZeRO + PP.
We recommend using Gemini to qucikly run your model in a distributed manner.
It doesn't require significant changes to the model structures, therefore you can apply it on a new model easily.
And use Titans as an advanced weapon to pursue a more extreme performance.
Titans has included the some typical models, such as Vit and GPT.
However, it requires some efforts to start if facing a new model structure.
### GeminiDPP/ZeRO + Tensor Parallelism ### GeminiDPP/ZeRO + Tensor Parallelism
```bash ```bash
@ -56,6 +62,11 @@ The `train_gpt_demo.py` provides three distributed plans, you can choose the pla
- Pytorch DDP - Pytorch DDP
- Pytorch ZeRO - Pytorch ZeRO
### Titans (Tensor Parallelism) + ZeRO + Pipeline Parallelism
Titans provides a customized GPT model, which uses distributed operators as building blocks.
In [./titans/README.md], we provide a hybrid parallelism of ZeRO, TP and PP.
You can switch parallel strategies using a config file.
## Performance ## Performance

View File

@ -5,7 +5,7 @@
You can download the preprocessed sample dataset for this demo via our [Google Drive sharing link](https://drive.google.com/file/d/1QKI6k-e2gJ7XgS8yIpgPPiMmwiBP_BPE/view?usp=sharing). You can download the preprocessed sample dataset for this demo via our [Google Drive sharing link](https://drive.google.com/file/d/1QKI6k-e2gJ7XgS8yIpgPPiMmwiBP_BPE/view?usp=sharing).
You can also avoid dataset preparation by using `--use_dummy_data` during running. You can also avoid dataset preparation by using `--use_dummy_dataset` during running.
## Run this Demo ## Run this Demo
@ -13,15 +13,15 @@ Use the following commands to install prerequisites.
```bash ```bash
# assuming using cuda 11.3 # assuming using cuda 11.3
conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3 -c pytorch pip install -r requirements.txt
pip install colossalai==0.1.9+torch1.11cu11.3 -f https://release.colossalai.org
``` ```
Use the following commands to execute training. Use the following commands to execute training.
```Bash ```Bash
#!/usr/bin/env sh #!/usr/bin/env sh
export DATA=/path/to/small-gpt-dataset.json' # if you want to use real dataset, then remove --use_dummy_dataset
# export DATA=/path/to/small-gpt-dataset.json'
# run on a single node # run on a single node
colossalai run --nproc_per_node=<num_gpus> train_gpt.py --config configs/<config_file> --from_torch colossalai run --nproc_per_node=<num_gpus> train_gpt.py --config configs/<config_file> --from_torch
@ -34,14 +34,14 @@ colossalai run --nproc_per_node=<num_gpus> \
train_gpt.py \ train_gpt.py \
--config configs/<config_file> \ --config configs/<config_file> \
--from_torch \ --from_torch \
--use_dummy_data --use_dummy_dataset
# run on multiple nodes with slurm # run on multiple nodes with slurm
srun python \ srun python \
train_gpt.py \ train_gpt.py \
--config configs/<config_file> \ --config configs/<config_file> \
--host <master_node> \ --host <master_node> \
--use_dummy_data --use_dummy_dataset
``` ```