RASMUS/Finnish-ASR-Canary-v2

1# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.2#3# Licensed under the Apache License, Version 2.0 (the "License");4# you may not use this file except in compliance with the License.5# You may obtain a copy of the License at6#7#     http://www.apache.org/licenses/LICENSE-2.08#9# Unless required by applicable law or agreed to in writing, software10# distributed under the License is distributed on an "AS IS" BASIS,11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.12# See the License for the specific language governing permissions and13# limitations under the License.14import os15 16import torch17from lightning.pytorch import Trainer18from omegaconf import OmegaConf19 20from nemo.collections.speechlm2 import DataModule, DuplexS2SDataset, DuplexS2SSpeechDecoderModel21from nemo.core.config import hydra_runner22from nemo.utils.exp_manager import exp_manager23from nemo.utils.trainer_utils import resolve_trainer_cfg24 25torch.cuda.set_device(int(os.environ["LOCAL_RANK"]))26 27 28@hydra_runner(config_path="conf", config_name="s2s_duplex_speech_decoder")29def train(cfg):30    OmegaConf.resolve(cfg)31    torch.distributed.init_process_group(backend="nccl")32    torch.set_float32_matmul_precision("medium")33    torch.backends.cudnn.allow_tf32 = True34    trainer = Trainer(**resolve_trainer_cfg(cfg.trainer))35    log_dir = exp_manager(trainer, cfg.get("exp_manager", None))36    OmegaConf.save(cfg, log_dir / "exp_config.yaml")37 38    with trainer.init_module():39        model = DuplexS2SSpeechDecoderModel(OmegaConf.to_container(cfg.model, resolve=True))40 41    dataset = DuplexS2SDataset(42        tokenizer=model.tokenizer,43        frame_length=cfg.data.frame_length,44        source_sample_rate=cfg.data.source_sample_rate,45        target_sample_rate=cfg.data.target_sample_rate,46        input_roles=cfg.data.input_roles,47        output_roles=cfg.data.output_roles,48    )49    datamodule = DataModule(cfg.data, tokenizer=model.tokenizer, dataset=dataset)50 51    trainer.fit(model, datamodule)52 53 54if __name__ == "__main__":55    train()56