Skip to content

[example] update bert example result #4458

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.

Sign up for GitHub

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

Merged
ver217 merged 1 commit into from
Aug 17, 2023
Merged

[example] update bert example result #4458

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
16 changes: 12 additions & 4 deletions examples/language/bert/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,16 +7,24 @@ This directory includes two parts: Using the Booster API finetune Huggingface Be
bash test_ci.sh
```

### Results on 2-GPU

| Plugin | Accuracy | F1-score |
| -------------- | -------- | -------- |
| torch_ddp | 84.4% | 88.6% |
| torch_ddp_fp16 | 84.7% | 88.8% |
| gemini | 84.0% | 88.4% |

## Benchmark
```
bash benchmark.sh
```

Now include these metrics in benchmark: CUDA mem occupy, throughput and the number of model parameters. If you have custom metrics, you can add them to benchmark_util.

## Results
### Results

### Bert
#### Bert

| | max cuda mem | throughput(sample/s) | params |
| :-----| -----------: | :--------: | :----: |
Expand All @@ -25,10 +33,10 @@ Now include these metrics in benchmark: CUDA mem occupy, throughput and the numb
| gemini | 11.0 GB | 12.9 | 82M |
| low_level_zero | 11.29 G | 14.7 | 82M |

### AlBert
#### AlBert
| | max cuda mem | throughput(sample/s) | params |
| :-----| -----------: | :--------: | :----: |
| ddp | OOM | | |
| ddp_fp16 | OOM | | |
| gemini | 69.39 G | 1.3 | 208M |
| low_level_zero | 56.89 G | 1.4 | 208M |
| low_level_zero | 56.89 G | 1.4 | 208M |
8 changes: 4 additions & 4 deletions examples/language/bert/finetune.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,8 +38,8 @@ def move_to_cuda(batch):


@torch.no_grad()
def evaluate_model(model: nn.Module, test_dataloader: Union[DataLoader, List[DataLoader]], num_labels: int, task_name: str,
eval_splits: List[str], coordinator: DistCoordinator):
def evaluate_model(model: nn.Module, test_dataloader: Union[DataLoader, List[DataLoader]], num_labels: int,
task_name: str, eval_splits: List[str], coordinator: DistCoordinator):
metric = evaluate.load("glue", task_name, process_id=coordinator.rank, num_process=coordinator.world_size)
model.eval()

Expand Down Expand Up @@ -142,7 +142,7 @@ def main():
if args.plugin.startswith('torch_ddp'):
plugin = TorchDDPPlugin()
elif args.plugin == 'gemini':
plugin = GeminiPlugin(placement_policy='cuda', strict_ddp_mode=True, initial_scale=2**5)
plugin = GeminiPlugin(initial_scale=2**5)
elif args.plugin == 'low_level_zero':
plugin = LowLevelZeroPlugin(initial_scale=2**5)

Expand Down Expand Up @@ -208,7 +208,7 @@ def main():
train_epoch(epoch, model, optimizer, lr_scheduler, train_dataloader, booster, coordinator)

results = evaluate_model(model, test_dataloader, data_builder.num_labels, args.task, data_builder.eval_splits,
coordinator)
coordinator)

if coordinator.is_master():
print(results)
Expand Down