wehub-resource-sync
ba4be087d5
Create PR to main with cherry-pick from release / cherry-pick (push) Failing after 0s
CICD NeMo / pre-flight (push) Failing after 0s
CICD NeMo / configure (push) Has been skipped
Build, validate, and release Neural Modules / pre-flight (push) Failing after 1s
CICD NeMo / code-linting (push) Has been skipped
Build, validate, and release Neural Modules / release (push) Has been skipped
Build, validate, and release Neural Modules / release-summary (push) Has been cancelled
CICD NeMo / cicd-test-container-build (push) Has been cancelled
CICD NeMo / cicd-import-tests (push) Has been cancelled
CICD NeMo / L0_Setup_Test_Data_And_Models (push) Has been cancelled
CICD NeMo / cicd-main-unit-tests (push) Has been cancelled
CICD NeMo / cicd-main-speech (push) Has been cancelled
CICD NeMo / Nemo_CICD_Test (push) Has been cancelled
CICD NeMo / Coverage (e2e) (push) Has been cancelled
CICD NeMo / Coverage (unit-test) (push) Has been cancelled
CodeQL / Analyze (python) (push) Has been cancelled
CICD NeMo / cicd-wait-in-queue (push) Has been cancelled
72 lines
2.6 KiB
Python
72 lines
2.6 KiB
Python
# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
import multiprocessing as mp
|
|
import os
|
|
|
|
import torch
|
|
from lightning.pytorch import Trainer
|
|
from lightning.pytorch.callbacks import ModelCheckpoint
|
|
from omegaconf import OmegaConf, open_dict
|
|
|
|
from nemo.collections.speechlm2 import DataModule, DuplexSTTDataset, DuplexSTTModel
|
|
from nemo.core.config import hydra_runner
|
|
from nemo.utils.exp_manager import exp_manager
|
|
from nemo.utils.trainer_utils import resolve_trainer_cfg
|
|
|
|
# Set multiprocessing start method to 'spawn' for CUDA compatibility with DataLoader workers
|
|
# This prevents "Cannot re-initialize CUDA in forked subprocess" errors
|
|
try:
|
|
mp.set_start_method('spawn', force=True)
|
|
except RuntimeError:
|
|
pass # Start method already set
|
|
|
|
torch.cuda.set_device(int(os.environ["LOCAL_RANK"]))
|
|
|
|
|
|
@hydra_runner(config_path="conf", config_name="duplex_stt")
|
|
def train(cfg):
|
|
OmegaConf.resolve(cfg)
|
|
torch.distributed.init_process_group(backend="nccl")
|
|
torch.set_float32_matmul_precision("medium")
|
|
torch.backends.cudnn.allow_tf32 = True
|
|
trainer = Trainer(**resolve_trainer_cfg(cfg.trainer))
|
|
log_dir = exp_manager(trainer, cfg.get("exp_manager", None))
|
|
OmegaConf.save(cfg, log_dir / "exp_config.yaml")
|
|
|
|
# avoid using `=` in the checkpoint name
|
|
for callback in trainer.callbacks:
|
|
if isinstance(callback, ModelCheckpoint):
|
|
callback.CHECKPOINT_EQUALS_CHAR = "-"
|
|
|
|
with trainer.init_module():
|
|
model = DuplexSTTModel(OmegaConf.to_container(cfg.model, resolve=True))
|
|
|
|
dataset = DuplexSTTDataset(
|
|
tokenizer=model.tokenizer,
|
|
frame_length=cfg.data.frame_length,
|
|
source_sample_rate=cfg.data.source_sample_rate,
|
|
input_roles=cfg.data.input_roles,
|
|
output_roles=cfg.data.output_roles,
|
|
aug_by_swap_role=cfg.data.get("aug_by_swap_role", False),
|
|
cfg=cfg.data,
|
|
model_cfg=cfg.model,
|
|
)
|
|
datamodule = DataModule(cfg.data, tokenizer=model.tokenizer, dataset=dataset)
|
|
|
|
trainer.fit(model, datamodule)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
train()
|