Skip to content
This repository has been archived by the owner on Aug 26, 2022. It is now read-only.

Commit

Permalink
Add troubleshooting
Browse files Browse the repository at this point in the history
  • Loading branch information
hyunwoongko committed Dec 29, 2021
1 parent b773493 commit 0fa0241
Showing 1 changed file with 11 additions and 4 deletions.
15 changes: 11 additions & 4 deletions USAGE.md
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
- [Deployment Launcher](#deployment-launcher)
- [Model Deployment](#model-deployment)
- [Docker Environment](#docker-environment)
- [Troubleshooting](#troubleshooting)

## 3D Parallelism

Expand Down Expand Up @@ -683,9 +684,9 @@ Note that you don't need to use distributed launcher when you use deployment lau

### Model Deployment

The usage of deployment launcher is similar with 3D parallelization.
The usage of deployment launcher is similar with 3D parallelism.
Just input the argument `deployment=True`.
Similar with 3D parallelism, the model can be created from split or merged checkpoints, and features such as kernel fusion can be used together.
The model can be created from split or merged checkpoints, and features such as kernel fusion can also be used together.

```python
from oslo import GPTNeoForCausalLM
Expand Down Expand Up @@ -756,5 +757,11 @@ But when deploying using pipeline parallelism, you don't have to write a new loo
So, please write your code as usual.

### Docker Environment
Deployment Launcher uses shared memory to share data between processes. However, Docker is designed to use limited shared memory by default. Therefore, when using the Deployment Launcher in a Docker container environment, the shared memory size must be increased, and the larger the model, the larger the shared memory is required.
You can set the larger shared memory size using `--shm-size=?gb`, and you can also disable shared memory limit by using `--ipc=host`.
Deployment Launcher uses shared memory to share data between processes. However, Docker is designed to use limited shared memory by default. Therefore, when using the Deployment Launcher in a Docker container, the shared memory size must be increased, and the larger the model, the larger the shared memory is required.
You can set the larger shared memory size using `--shm-size=?gb`, and you can also disable shared memory limit by using `--ipc=host`.

## Troubleshooting
If you have any questions, bug reports, and feature requests, please open an issue on github or
contacts [contact@tunib.ai](mailto:contact@tunib.ai) please.

We appreciate any kind of feedback or contribution. Feel free to proceed with small issues like bug fixes, documentation improvement. For major contributions and new features, please discuss with the collaborators in corresponding issues.

0 comments on commit 0fa0241

Please sign in to comment.