diff --git a/src/nemotron/cli/commands/embed/finetune.py b/src/nemotron/cli/commands/embed/finetune.py index b9e1ca241..427c52c13 100644 --- a/src/nemotron/cli/commands/embed/finetune.py +++ b/src/nemotron/cli/commands/embed/finetune.py @@ -115,7 +115,9 @@ def _execute_uv_local(train_path: Path, passthrough: list[str]) -> None: uv_cmd, "run", "--with", str(repo_root), "--project", str(stage_dir), - "python", str(script_abs), + "python", "-m", "torch.distributed.run", + "--nproc_per_node=gpu", + str(script_abs), "--config", str(train_path), *passthrough, ] diff --git a/src/nemotron/recipes/embed/stage2_finetune/train.py b/src/nemotron/recipes/embed/stage2_finetune/train.py index b2e095d5f..88f1f59ac 100644 --- a/src/nemotron/recipes/embed/stage2_finetune/train.py +++ b/src/nemotron/recipes/embed/stage2_finetune/train.py @@ -8,7 +8,7 @@ # setup = "PyTorch pre-installed. Stage dependencies resolved via UV at runtime." # # [tool.runspec.run] -# launch = "direct" +# launch = "torchrun" # # [tool.runspec.config] # dir = "./config"