Skip to content

separate out training for fault tolerance #2311

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
tushar00jain merged 1 commit into from
Feb 11, 2026
+524 −45
Merged

separate out training for fault tolerance #2311

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
2 changes: 2 additions & 0 deletions tests/integration_tests/ft.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,9 @@ def run_single_test(test_flavor: OverrideDefinitions, full_path: str, output_dir
cmd = (
f'TORCH_TRACE="{output_dir}/{test_name}/compile_trace" '
+ f"CUDA_VISIBLE_DEVICES={ranks} "
+ "TRAIN_FILE=torchtitan.experiments.ft.train "
+ f"CONFIG_FILE={full_path} NGPU={test_flavor.ngpu} ./run_train.sh "
+ "--model.name=llama3_ft "
+ "--fault_tolerance.enable "
+ f"--fault_tolerance.replica_id={replica_id} --fault_tolerance.group_size={test_flavor.ngpu}"
)
Expand Down
8 changes: 4 additions & 4 deletions docs/torchft.md → torchtitan/experiments/ft/torchft.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,12 +36,12 @@ RUST_BACKTRACE=1 torchft_lighthouse --min_replicas 1 --quorum_tick_ms 100 --join
2. Launch the first TorchTitan instance:

```bash
NGPU=4 CUDA_VISIBLE_DEVICES=0,1,2,3 CONFIG_FILE="./torchtitan/models/llama3/train_configs/llama3_8b.toml" ./run_train.sh --fault_tolerance.enable --fault_tolerance.replica_id=0 --fault_tolerance.group_size=2 --parallelism.data_parallel_shard_degree=4
NGPU=4 CUDA_VISIBLE_DEVICES=0,1,2,3 TRAIN_FILE=torchtitan.experiments.ft.train CONFIG_FILE="./torchtitan/models/llama3/train_configs/llama3_8b.toml" ./run_train.sh --fault_tolerance.enable --fault_tolerance.replica_id=0 --fault_tolerance.group_size=2 --parallelism.data_parallel_shard_degree=4
```
3. Launch the second TorchTitan instance:

```bash
NGPU=4 CUDA_VISIBLE_DEVICES=4,5,6,7 CONFIG_FILE="./torchtitan/models/llama3/train_configs/llama3_8b.toml" ./run_train.sh --fault_tolerance.enable --fault_tolerance.replica_id=1 --fault_tolerance.group_size=2 --parallelism.data_parallel_shard_degree=4
NGPU=4 CUDA_VISIBLE_DEVICES=4,5,6,7 TRAIN_FILE=torchtitan.experiments.ft.train CONFIG_FILE="./torchtitan/models/llama3/train_configs/llama3_8b.toml" ./run_train.sh --fault_tolerance.enable --fault_tolerance.replica_id=1 --fault_tolerance.group_size=2 --parallelism.data_parallel_shard_degree=4
```

### Explanation
Expand All @@ -68,12 +68,12 @@ The `--training.global_batch_size` parameter refers to global batch size that wi

#### Replica Group 0
```bash
CONFIG_FILE=./torchtitan/models/llama3_ft/train_configs/debug_model.toml CUDA_VISIBLE_DEVICES=0,1,2,3 NGPU=4 ./run_train.sh --parallelism.data_parallel_shard_degree=4 --fault_tolerance.enable --fault_tolerance.group_size=2 --fault_tolerance.replica_id=0
TRAIN_FILE=torchtitan.experiments.ft.train CONFIG_FILE=./torchtitan/models/llama3_ft/train_configs/debug_model.toml CUDA_VISIBLE_DEVICES=0,1,2,3 NGPU=4 ./run_train.sh --parallelism.data_parallel_shard_degree=4 --fault_tolerance.enable --fault_tolerance.group_size=2 --fault_tolerance.replica_id=0
```

#### Replica Group 1
```bash
CONFIG_FILE=./torchtitan/models/llama3_ft/train_configs/debug_model.toml CUDA_VISIBLE_DEVICES=4,5,6,7 NGPU=4 ./run_train.sh --parallelism.data_parallel_shard_degree=4 --fault_tolerance.enable --fault_tolerance.group_size=2 --fault_tolerance.replica_id=1
TRAIN_FILE=torchtitan.experiments.ft.train CONFIG_FILE=./torchtitan/models/llama3_ft/train_configs/debug_model.toml CUDA_VISIBLE_DEVICES=4,5,6,7 NGPU=4 ./run_train.sh --parallelism.data_parallel_shard_degree=4 --fault_tolerance.enable --fault_tolerance.group_size=2 --fault_tolerance.replica_id=1
```

## Fault Tolerance Configuration Options
Expand Down
Loading
Loading