Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

241 Table rendering in Chapter 7 is messed up #249

Merged
merged 1 commit into from
Jun 4, 2024
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 13 additions & 13 deletions contents/training/training.qmd
Original file line number Diff line number Diff line change
Expand Up @@ -1010,19 +1010,19 @@ However, as the model parts run on physically separate devices, they must commun

### Comparison

To summarize, @tbl-parallelism demonstrates some of the key characteristics for comparing data parallelism and model parallelism:

| Characteristic | Data Parallelism | Model Parallelism |
|-|-----------------|-------------------|
| Definition | Distribute data across devices with model replicas | Distribute model across devices |
| Objective | Accelerate training through compute scaling | Enable larger model training |
| Scaling Method | Scale devices/workers | Scale model size|
| Main Constraint | Model size per device | Device coordination overhead |
| Hardware Requirements | Multiple GPU/TPUs | Often specialized interconnect |
| Primary Challenge | Parameter synchronization | Complex partitioning + communication |
| Types | N/A | Layer-wise, filter-wise, spatial |
| Code Complexity | Minimal changes | More significant model surgery |
| Popular Libraries | Horovod, PyTorch Distributed | Mesh TensorFlow |
To summarize, `@tbl-parallelism` demonstrates some of the key characteristics for comparing data parallelism and model parallelism:

| Characteristic | Data Parallelism | Model Parallelism |
|----------------------|----------------------------------------------------|--------------------------------------|
| Definition | Distribute data across devices with model replicas | Distribute model across devices |
| Objective | Accelerate training through compute scaling | Enable larger model training |
| Scaling Method | Scale devices/workers | Scale model size |
| Main Constraint | Model size per device | Device coordination overhead |
| Hardware Requirements| Multiple GPU/TPUs | Often specialized interconnect |
| Primary Challenge | Parameter synchronization | Complex partitioning + communication |
| Types | N/A | Layer-wise, filter-wise, spatial |
| Code Complexity | Minimal changes | More significant model surgery |
| Popular Libraries | Horovod, PyTorch Distributed | Mesh TensorFlow |

: Comparing data parallelism and model parallelism. {#tbl-parallelism}

Expand Down
Loading