Merge pull request #983 from Shengqiang-Li/main

feat: Support DPO

cosyvoice/bin/train_dpo.py

@@ -0,0 +1,187 @@
+# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu)
+#
+# 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.
+
+from __future__ import print_function
+import argparse
+import datetime
+import logging
+logging.getLogger('matplotlib').setLevel(logging.WARNING)
+from copy import deepcopy
+import os
+import torch
+import torch.distributed as dist
+import deepspeed
+
+from hyperpyyaml import load_hyperpyyaml
+
+from torch.distributed.elastic.multiprocessing.errors import record
+
+from cosyvoice.utils.executor_dpo import Executor
+from cosyvoice.utils.train_utils_dpo import (
+    init_distributed,
+    init_dataset_and_dataloader,
+    init_optimizer_and_scheduler,
+    init_summarywriter, save_model,
+    wrap_cuda_model, check_modify_and_save_config)
+
+
+def get_args():
+    parser = argparse.ArgumentParser(description='training your network')
+    parser.add_argument('--train_engine',
+                        default='torch_ddp',
+                        choices=['torch_ddp', 'deepspeed'],
+                        help='Engine for paralleled training')
+    parser.add_argument('--model', required=True, help='model which will be trained')
+    parser.add_argument('--config', required=True, help='config file')
+    parser.add_argument('--train_data', required=True, help='train data file')
+    parser.add_argument('--cv_data', required=True, help='cv data file')
+    parser.add_argument('--checkpoint', help='checkpoint model')
+    parser.add_argument('--model_dir', required=True, help='save model dir')
+    parser.add_argument('--tensorboard_dir',
+                        default='tensorboard',
+                        help='tensorboard log dir')
+    parser.add_argument('--ddp.dist_backend',
+                        dest='dist_backend',
+                        default='nccl',
+                        choices=['nccl', 'gloo'],
+                        help='distributed backend')
+    parser.add_argument('--num_workers',
+                        default=0,
+                        type=int,
+                        help='num of subprocess workers for reading')
+    parser.add_argument('--prefetch',
+                        default=100,
+                        type=int,
+                        help='prefetch number')
+    parser.add_argument('--pin_memory',
+                        action='store_true',
+                        default=False,
+                        help='Use pinned memory buffers used for reading')
+    parser.add_argument('--use_amp',
+                        action='store_true',
+                        default=False,
+                        help='Use automatic mixed precision training')
+    parser.add_argument('--deepspeed.save_states',
+                        dest='save_states',
+                        default='model_only',
+                        choices=['model_only', 'model+optimizer'],
+                        help='save model/optimizer states')
+    parser.add_argument('--timeout',
+                        default=60,
+                        type=int,
+                        help='timeout (in seconds) of cosyvoice_join.')
+    parser.add_argument('--dpo',
+                        action='store_true',
+                        default=False,
+                        help='Use Direct Preference Optimization')
+    parser.add_argument('--beta',
+                        default=0.01,
+                        type=float,
+                        help='beta of dpo training')
+    parser = deepspeed.add_config_arguments(parser)
+    args = parser.parse_args()
+    return args
+
+
+@record
+def main():
+    args = get_args()
+    logging.basicConfig(level=logging.DEBUG,
+                        format='%(asctime)s %(levelname)s %(message)s')
+    # gan train has some special initialization logic
+    gan = True if args.model == 'hifigan' else False
+
+    override_dict = {k: None for k in ['llm', 'flow', 'hift', 'hifigan'] if k != args.model}
+    if gan is True:
+        override_dict.pop('hift')
+    with open(args.config, 'r') as f:
+        configs = load_hyperpyyaml(f, overrides=override_dict)
+    if gan is True:
+        configs['train_conf'] = configs['train_conf_gan']
+    configs['train_conf'].update(vars(args))
+
+    # Init env for ddp
+    init_distributed(args)
+
+    # Get dataset & dataloader
+    train_dataset, cv_dataset, train_data_loader, cv_data_loader = \
+        init_dataset_and_dataloader(args, configs, gan)
+
+    # Do some sanity checks and save config to arsg.model_dir
+    configs = check_modify_and_save_config(args, configs)
+
+    # Tensorboard summary
+    writer = init_summarywriter(args)
+
+    # load checkpoint
+    model = configs[args.model]
+    ref_model = None
+    if args.dpo:
+        ref_model = deepcopy(model)
+    start_step, start_epoch = 0, -1
+    if args.checkpoint is not None:
+        if os.path.exists(args.checkpoint):
+            state_dict = torch.load(args.checkpoint, map_location='cpu')
+            model.load_state_dict(state_dict, strict=False)
+            if args.dpo:
+                ref_model.load_state_dict(state_dict, strict=False)
+            if 'step' in state_dict:
+                start_step = state_dict['step']
+            if 'epoch' in state_dict:
+                start_epoch = state_dict['epoch']
+        else:
+            logging.warning('checkpoint {} do not exsist!'.format(args.checkpoint))
+
+    # Dispatch model from cpu to gpu
+    model = wrap_cuda_model(args, model)
+    if args.dpo:
+        ref_model = wrap_cuda_model(args, ref_model)
+
+    # Get optimizer & scheduler
+    model, optimizer, scheduler, optimizer_d, scheduler_d = init_optimizer_and_scheduler(args, configs, model, gan)
+    if args.dpo:
+        ref_model, _, _, _, _ = init_optimizer_and_scheduler(args, configs, ref_model, gan)
+    scheduler.set_step(start_step)
+    if scheduler_d is not None:
+        scheduler_d.set_step(start_step)
+
+    # Save init checkpoints
+    info_dict = deepcopy(configs['train_conf'])
+    info_dict['step'] = start_step
+    info_dict['epoch'] = start_epoch
+    save_model(model, 'init', info_dict)
+
+    # Get executor
+    executor = Executor(gan=gan, dpo=args.dpo, beta=args.beta)
+    executor.step = start_step
+
+    # Init scaler, used for pytorch amp mixed precision training
+    scaler = torch.cuda.amp.GradScaler() if args.use_amp else None
+    print('start step {} start epoch {}'.format(start_step, start_epoch))
+    # Start training loop
+    for epoch in range(start_epoch + 1, info_dict['max_epoch']):
+        executor.epoch = epoch
+        train_dataset.set_epoch(epoch)
+        dist.barrier()
+        group_join = dist.new_group(backend="gloo", timeout=datetime.timedelta(seconds=args.timeout))
+        if gan is True:
+            executor.train_one_epoc_gan(model, optimizer, scheduler, optimizer_d, scheduler_d, train_data_loader, cv_data_loader,
+                                        writer, info_dict, scaler, group_join)
+        else:
+            executor.train_one_epoc(model, optimizer, scheduler, train_data_loader, cv_data_loader, writer, info_dict, scaler, group_join, ref_model)
+        dist.destroy_process_group(group_join)
+
+
+if __name__ == '__main__':
+    main()

cosyvoice/dataset/processor_dpo.py

@@ -0,0 +1,443 @@
+# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu)
+#
+# 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 logging
+import random
+
+import pyarrow.parquet as pq
+from io import BytesIO
+import torch
+import torchaudio
+from torch.nn.utils.rnn import pad_sequence
+import torch.nn.functional as F
+import pyworld as pw
+
+
+AUDIO_FORMAT_SETS = {'flac', 'mp3', 'm4a', 'ogg', 'opus', 'wav', 'wma'}
+
+
+def parquet_opener(data, mode='train', tts_data={}):
+    """ Give url or local file, return file descriptor
+        Inplace operation.
+
+        Args:
+            data(Iterable[str]): url or local file list
+
+        Returns:
+            Iterable[{src, stream}]
+    """
+    for sample in data:
+        assert 'src' in sample
+        url = sample['src']
+        try:
+            for df in pq.ParquetFile(url).iter_batches(batch_size=64):
+                df = df.to_pandas()
+                for i in range(len(df)):
+                    if mode == 'inference' and df.loc[i, 'utt'] not in tts_data:
+                        continue
+                    sample.update(dict(df.loc[i]))
+                    if mode == 'train':
+                        # NOTE do not return sample directly, must initialize a new dict
+                        yield {**sample}
+                    else:
+                        for index, text in enumerate(tts_data[df.loc[i, 'utt']]):
+                            yield {**sample, 'tts_index': index, 'tts_text': text}
+        except Exception as ex:
+            logging.warning('Failed to open {}, ex info {}'.format(url, ex))
+
+
+def filter(data,
+           max_length=10240,
+           min_length=10,
+           token_max_length=200,
+           token_min_length=1,
+           min_output_input_ratio=0.0005,
+           max_output_input_ratio=1,
+           mode='train'):
+    """ Filter sample according to feature and label length
+        Inplace operation.
+
+        Args::
+            data: Iterable[{key, wav, label, sample_rate}]
+            max_length: drop utterance which is greater than max_length(10ms)
+            min_length: drop utterance which is less than min_length(10ms)
+            token_max_length: drop utterance which is greater than
+                token_max_length, especially when use char unit for
+                english modeling
+            token_min_length: drop utterance which is
+                less than token_max_length
+            min_output_input_ratio: minimal ration of
+                token_length / feats_length(10ms)
+            max_output_input_ratio: maximum ration of
+                token_length / feats_length(10ms)
+
+        Returns:
+            Iterable[{key, wav, label, sample_rate}]
+    """
+    for sample in data:
+        sample['speech'], sample['sample_rate'] = torchaudio.load(BytesIO(sample['audio_data']))
+        sample['speech'] = sample['speech'].mean(dim=0, keepdim=True)
+        del sample['audio_data']
+        # sample['wav'] is torch.Tensor, we have 100 frames every second
+        num_frames = sample['speech'].size(1) / sample['sample_rate'] * 100
+        if num_frames < min_length:
+            continue
+        if num_frames > max_length:
+            continue
+        if len(sample['text_token']) < token_min_length:
+            continue
+        if len(sample['text_token']) > token_max_length:
+            continue
+        if len(sample['speech_token']) == 0:
+            continue
+        if num_frames != 0:
+            if len(sample['text_token']) / num_frames < min_output_input_ratio:
+                continue
+            if len(sample['text_token']) / num_frames > max_output_input_ratio:
+                continue
+        yield sample
+
+
+def resample(data, resample_rate=22050, min_sample_rate=16000, mode='train'):
+    """ Resample data.
+        Inplace operation.
+
+        Args:
+            data: Iterable[{key, wav, label, sample_rate}]
+            resample_rate: target resample rate
+
+        Returns:
+            Iterable[{key, wav, label, sample_rate}]
+    """
+    for sample in data:
+        assert 'sample_rate' in sample
+        assert 'speech' in sample
+        sample_rate = sample['sample_rate']
+        waveform = sample['speech']
+        if sample_rate != resample_rate:
+            if sample_rate < min_sample_rate:
+                continue
+            sample['sample_rate'] = resample_rate
+            sample['speech'] = torchaudio.transforms.Resample(
+                orig_freq=sample_rate, new_freq=resample_rate)(waveform)
+        max_val = sample['speech'].abs().max()
+        if max_val > 1:
+            sample['speech'] /= max_val
+        yield sample
+
+
+def truncate(data, truncate_length=24576, mode='train'):
+    """ Truncate data.
+
+        Args:
+            data: Iterable[{key, wav, label, sample_rate}]
+            truncate_length: truncate length
+
+        Returns:
+            Iterable[{key, wav, label, sample_rate}]
+    """
+    for sample in data:
+        waveform = sample['speech']
+        if waveform.shape[1] > truncate_length:
+            start = random.randint(0, waveform.shape[1] - truncate_length)
+            waveform = waveform[:, start: start + truncate_length]
+        else:
+            waveform = torch.concat([waveform, torch.zeros(1, truncate_length - waveform.shape[1])], dim=1)
+        sample['speech'] = waveform
+        yield sample
+
+
+def compute_fbank(data,
+                  feat_extractor,
+                  mode='train'):
+    """ Extract fbank
+
+        Args:
+            data: Iterable[{key, wav, label, sample_rate}]
+
+        Returns:
+            Iterable[{key, feat, label}]
+    """
+    for sample in data:
+        assert 'sample_rate' in sample
+        assert 'speech' in sample
+        assert 'utt' in sample
+        assert 'text_token' in sample
+        waveform = sample['speech']
+        mat = feat_extractor(waveform).squeeze(dim=0).transpose(0, 1)
+        sample['speech_feat'] = mat
+        yield sample
+
+
+def compute_f0(data, sample_rate, hop_size, mode='train'):
+    """ Extract f0
+
+        Args:
+            data: Iterable[{key, wav, label, sample_rate}]
+
+        Returns:
+            Iterable[{key, feat, label}]
+    """
+    frame_period = hop_size * 1000 / sample_rate
+    for sample in data:
+        assert 'sample_rate' in sample
+        assert 'speech' in sample
+        assert 'utt' in sample
+        assert 'text_token' in sample
+        waveform = sample['speech']
+        _f0, t = pw.harvest(waveform.squeeze(dim=0).numpy().astype('double'), sample_rate, frame_period=frame_period)
+        if sum(_f0 != 0) < 5: # this happens when the algorithm fails
+            _f0, t = pw.dio(waveform.squeeze(dim=0).numpy().astype('double'), sample_rate, frame_period=frame_period) # if harvest fails, try dio
+        f0 = pw.stonemask(waveform.squeeze(dim=0).numpy().astype('double'), _f0, t, sample_rate)
+        f0 = F.interpolate(torch.from_numpy(f0).view(1, 1, -1), size=sample['speech_feat'].shape[0], mode='linear').view(-1)
+        sample['pitch_feat'] = f0
+        yield sample
+
+
+def parse_embedding(data, normalize, mode='train'):
+    """ Parse utt_embedding/spk_embedding
+
+        Args:
+            data: Iterable[{key, wav, label, sample_rate}]
+
+        Returns:
+            Iterable[{key, feat, label}]
+    """
+    for sample in data:
+        sample['utt_embedding'] = torch.tensor(sample['utt_embedding'], dtype=torch.float32)
+        sample['spk_embedding'] = torch.tensor(sample['spk_embedding'], dtype=torch.float32)
+        if normalize:
+            sample['utt_embedding'] = F.normalize(sample['utt_embedding'], dim=0)
+            sample['spk_embedding'] = F.normalize(sample['spk_embedding'], dim=0)
+        yield sample
+
+
+def tokenize(data, get_tokenizer, allowed_special, mode='train'):
+    """ Decode text to chars or BPE
+        Inplace operation
+
+        Args:
+            data: Iterable[{key, wav, txt, sample_rate}]
+
+        Returns:
+            Iterable[{key, wav, txt, tokens, label, sample_rate}]
+    """
+    tokenizer = get_tokenizer()
+    for sample in data:
+        assert 'text' in sample
+        sample['text_token'] = tokenizer.encode(sample['text'], allowed_special=allowed_special)
+        if mode == 'inference':
+            sample['tts_text_token'] = tokenizer.encode(sample['tts_text'], allowed_special=allowed_special)
+        yield sample
+
+
+def shuffle(data, shuffle_size=10000, mode='train'):
+    """ Local shuffle the data
+
+        Args:
+            data: Iterable[{key, feat, label}]
+            shuffle_size: buffer size for shuffle
+
+        Returns:
+            Iterable[{key, feat, label}]
+    """
+    buf = []
+    for sample in data:
+        buf.append(sample)
+        if len(buf) >= shuffle_size:
+            random.shuffle(buf)
+            for x in buf:
+                yield x
+            buf = []
+    # The sample left over
+    random.shuffle(buf)
+    for x in buf:
+        yield x
+
+
+def sort(data, sort_size=500, mode='train'):
+    """ Sort the data by feature length.
+        Sort is used after shuffle and before batch, so we can group
+        utts with similar lengths into a batch, and `sort_size` should
+        be less than `shuffle_size`
+
+        Args:
+            data: Iterable[{key, feat, label}]
+            sort_size: buffer size for sort
+
+        Returns:
+            Iterable[{key, feat, label}]
+    """
+
+    buf = []
+    for sample in data:
+        buf.append(sample)
+        if len(buf) >= sort_size:
+            buf.sort(key=lambda x: x['speech_feat'].size(0))
+            for x in buf:
+                yield x
+            buf = []
+    # The sample left over
+    buf.sort(key=lambda x: x['speech_feat'].size(0))
+    for x in buf:
+        yield x
+
+
+def static_batch(data, batch_size=16):
+    """ Static batch the data by `batch_size`
+
+        Args:
+            data: Iterable[{key, feat, label}]
+            batch_size: batch size
+
+        Returns:
+            Iterable[List[{key, feat, label}]]
+    """
+    buf = []
+    for sample in data:
+        buf.append(sample)
+        if len(buf) >= batch_size:
+            yield buf
+            buf = []
+    if len(buf) > 0:
+        yield buf
+
+
+def dynamic_batch(data, max_frames_in_batch=12000, mode='train'):
+    """ Dynamic batch the data until the total frames in batch
+        reach `max_frames_in_batch`
+
+        Args:
+            data: Iterable[{key, feat, label}]
+            max_frames_in_batch: max_frames in one batch
+
+        Returns:
+            Iterable[List[{key, feat, label}]]
+    """
+    buf = []
+    longest_frames = 0
+    for sample in data:
+        assert 'speech_feat' in sample
+        assert isinstance(sample['speech_feat'], torch.Tensor)
+        new_sample_frames = sample['speech_feat'].size(0)
+        longest_frames = max(longest_frames, new_sample_frames)
+        frames_after_padding = longest_frames * (len(buf) + 1)
+        if frames_after_padding > max_frames_in_batch:
+            yield buf
+            buf = [sample]
+            longest_frames = new_sample_frames
+        else:
+            buf.append(sample)
+    if len(buf) > 0:
+        yield buf
+
+
+def batch(data, batch_type='static', batch_size=16, max_frames_in_batch=12000, mode='train'):
+    """ Wrapper for static/dynamic batch
+    """
+    if mode == 'inference':
+        return static_batch(data, 1)
+    else:
+        if batch_type == 'static':
+            return static_batch(data, batch_size)
+        elif batch_type == 'dynamic':
+            return dynamic_batch(data, max_frames_in_batch)
+        else:
+            logging.fatal('Unsupported batch type {}'.format(batch_type))
+
+
+def padding(data, use_spk_embedding, mode='train', gan=False, dpo=False):
+    """ Padding the data into training data
+
+        Args:
+            data: Iterable[List[{key, feat, label}]]
+
+        Returns:
+            Iterable[Tuple(keys, feats, labels, feats lengths, label lengths)]
+    """
+    for sample in data:
+        assert isinstance(sample, list)
+        speech_feat_len = torch.tensor([x['speech_feat'].size(1) for x in sample],
+                                       dtype=torch.int32)
+        order = torch.argsort(speech_feat_len, descending=True)
+
+        utts = [sample[i]['utt'] for i in order]
+        speech = [sample[i]['speech'].squeeze(dim=0) for i in order]
+        speech_len = torch.tensor([i.size(0) for i in speech], dtype=torch.int32)
+        speech = pad_sequence(speech, batch_first=True, padding_value=0)
+        speech_token = [torch.tensor(sample[i]['speech_token']) for i in order]
+        speech_token_len = torch.tensor([i.size(0) for i in speech_token], dtype=torch.int32)
+        speech_token = pad_sequence(speech_token,
+                                    batch_first=True,
+                                    padding_value=0)
+        speech_feat = [sample[i]['speech_feat'] for i in order]
+        speech_feat_len = torch.tensor([i.size(0) for i in speech_feat], dtype=torch.int32)
+        speech_feat = pad_sequence(speech_feat,
+                                   batch_first=True,
+                                   padding_value=0)
+        text = [sample[i]['text'] for i in order]
+        text_token = [torch.tensor(sample[i]['text_token']) for i in order]
+        text_token_len = torch.tensor([i.size(0) for i in text_token], dtype=torch.int32)
+        text_token = pad_sequence(text_token, batch_first=True, padding_value=0)
+        utt_embedding = torch.stack([sample[i]['utt_embedding'] for i in order], dim=0)
+        spk_embedding = torch.stack([sample[i]['spk_embedding'] for i in order], dim=0)
+        batch = {
+            "utts": utts,
+            "speech": speech,
+            "speech_len": speech_len,
+            "speech_token": speech_token,
+            "speech_token_len": speech_token_len,
+            "speech_feat": speech_feat,
+            "speech_feat_len": speech_feat_len,
+            "text": text,
+            "text_token": text_token,
+            "text_token_len": text_token_len,
+            "utt_embedding": utt_embedding,
+            "spk_embedding": spk_embedding,
+        }
+        if dpo:
+            reject_speech_token = [torch.tensor(sample[i]['reject_speech_token']) for i in order]
+            reject_speech_token_len = torch.tensor([i.size(0) for i in reject_speech_token], dtype=torch.int32)
+            reject_speech_token = pad_sequence(reject_speech_token,
+                                                batch_first=True,
+                                                padding_value=0)
+            batch['reject_speech_token'] = reject_speech_token
+            batch['reject_speech_token_len'] = reject_speech_token_len
+        if gan is True:
+            # in gan train, we need pitch_feat
+            pitch_feat = [sample[i]['pitch_feat'] for i in order]
+            pitch_feat_len = torch.tensor([i.size(0) for i in pitch_feat], dtype=torch.int32)
+            pitch_feat = pad_sequence(pitch_feat,
+                                      batch_first=True,
+                                      padding_value=0)
+            batch["pitch_feat"] = pitch_feat
+            batch["pitch_feat_len"] = pitch_feat_len
+        else:
+            # only gan train needs speech, delete it to save memory
+            del batch["speech"]
+            del batch["speech_len"]
+        if mode == 'inference':
+            tts_text = [sample[i]['tts_text'] for i in order]
+            tts_index = [sample[i]['tts_index'] for i in order]
+            tts_text_token = [torch.tensor(sample[i]['tts_text_token']) for i in order]
+            tts_text_token_len = torch.tensor([i.size(0) for i in tts_text_token], dtype=torch.int32)
+            tts_text_token = pad_sequence(tts_text_token, batch_first=True, padding_value=-1)
+            batch.update({'tts_text': tts_text,
+                          'tts_index': tts_index,
+                          'tts_text_token': tts_text_token,
+                          'tts_text_token_len': tts_text_token_len})
+        if use_spk_embedding is True:
+            batch["embedding"] = batch["spk_embedding"]
+        else:
+            batch["embedding"] = batch["utt_embedding"]
+        yield batch

cosyvoice/llm/llm_dpo.py

@@ -0,0 +1,556 @@
+# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu, Zhihao Du)
+#
+# 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.
+from typing import Dict, Optional, Callable, List, Generator
+import torch
+from torch import nn
+import torch.nn.functional as F
+from transformers import Qwen2ForCausalLM
+from torch.nn.utils.rnn import pad_sequence, unpad_sequence
+from cosyvoice.utils.common import IGNORE_ID
+from cosyvoice.transformer.label_smoothing_loss import LabelSmoothingLoss
+from cosyvoice.utils.common import th_accuracy
+from cosyvoice.utils.file_utils import logging
+from cosyvoice.utils.mask import make_pad_mask
+
+
+class TransformerLM(torch.nn.Module):
+    def __init__(
+            self,
+            text_encoder_input_size: int,
+            llm_input_size: int,
+            llm_output_size: int,
+            text_token_size: int,
+            speech_token_size: int,
+            text_encoder: torch.nn.Module,
+            llm: torch.nn.Module,
+            sampling: Callable,
+            length_normalized_loss: bool = True,
+            lsm_weight: float = 0.0,
+            spk_embed_dim: int = 192,
+    ):
+        super().__init__()
+        self.llm_input_size = llm_input_size
+        self.speech_token_size = speech_token_size
+        # 1. build text token inputs related modules
+        self.text_embedding = torch.nn.Embedding(text_token_size, text_encoder_input_size)
+        self.text_encoder = text_encoder
+        self.text_encoder_affine_layer = nn.Linear(
+            self.text_encoder.output_size(),
+            llm_input_size
+        )
+
+        # 2. build speech token language model related modules
+        self.sos_eos = 0
+        self.task_id = 1
+        self.llm_embedding = torch.nn.Embedding(2, llm_input_size)
+        self.llm = llm
+        self.llm_decoder = nn.Linear(llm_output_size, speech_token_size + 1)
+        self.criterion_ce = LabelSmoothingLoss(
+            size=speech_token_size + 1,
+            padding_idx=IGNORE_ID,
+            smoothing=lsm_weight,
+            normalize_length=length_normalized_loss,
+        )
+
+        # 3. [Optional] build speech token related modules
+        self.speech_embedding = torch.nn.Embedding(speech_token_size, llm_input_size)
+        self.spk_embed_affine_layer = torch.nn.Linear(spk_embed_dim, llm_input_size)
+
+        # 4. sampling method
+        self.sampling = sampling
+
+    def encode(
+            self,
+            text: torch.Tensor,
+            text_lengths: torch.Tensor,
+    ):
+        encoder_out, encoder_mask = self.text_encoder(text, text_lengths, decoding_chunk_size=1, num_decoding_left_chunks=-1)
+        encoder_out_lens = encoder_mask.squeeze(1).sum(1)
+        encoder_out = self.text_encoder_affine_layer(encoder_out)
+        return encoder_out, encoder_out_lens
+
+    def pad_unpad_sequence(self, sos_eos_emb, embedding, text_token, text_token_len, task_id_emb, speech_token, speech_token_len):
+        text_token = unpad_sequence(text_token, text_token_len.cpu(), batch_first=True)
+        speech_token = unpad_sequence(speech_token, speech_token_len.cpu(), batch_first=True)
+        lm_input = [torch.concat([sos_eos_emb.squeeze(dim=0), embedding[i], text_token[i], task_id_emb.squeeze(dim=0), speech_token[i]], dim=0)
+                    for i in range(len(text_token))]
+        lm_input_len = torch.tensor([i.size(0) for i in lm_input], dtype=torch.int32)
+        lm_input = pad_sequence(lm_input, batch_first=True, padding_value=IGNORE_ID)
+        return lm_input, lm_input_len
+
+    def forward(
+            self,
+            batch: dict,
+            device: torch.device,
+    ) -> Dict[str, Optional[torch.Tensor]]:
+        """
+        Args:
+            text: (B, L, D)
+            text_lengths: (B,)
+            audio: (B, T, N) or (B, T)
+            audio_lengths: (B,)
+        """
+        text_token = batch['text_token'].to(device)
+        text_token_len = batch['text_token_len'].to(device)
+        speech_token = batch['speech_token'].to(device)
+        speech_token_len = batch['speech_token_len'].to(device)
+        embedding = batch['embedding'].to(device)
+
+        # 1. prepare llm_target
+        lm_target = [torch.tensor([IGNORE_ID] * (2 + text_token_len[i]) + speech_token[i, :speech_token_len[i]].tolist() +
+                                  [self.speech_token_size]) for i in range(text_token.size(0))]
+        lm_target = pad_sequence(lm_target, batch_first=True, padding_value=IGNORE_ID).to(device)
+
+        # 1. encode text_token
+        text_token = self.text_embedding(text_token)
+        text_token, text_token_len = self.encode(text_token, text_token_len)
+
+        # 2. embedding projection
+        embedding = F.normalize(embedding, dim=1)
+        embedding = self.spk_embed_affine_layer(embedding)
+        embedding = embedding.unsqueeze(1)
+
+        # 3. eos and task_id
+        sos_eos_emb = self.llm_embedding.weight[self.sos_eos].reshape(1, 1, -1)
+        task_id_emb = self.llm_embedding.weight[self.task_id].reshape(1, 1, -1)
+
+        # 4. encode speech_token
+        speech_token = self.speech_embedding(speech_token)
+
+        # 5. unpad and pad
+        lm_input, lm_input_len = self.pad_unpad_sequence(sos_eos_emb, embedding, text_token, text_token_len,
+                                                         task_id_emb, speech_token, speech_token_len)
+
+        # 6. run lm forward
+        lm_output, lm_output_mask = self.llm(lm_input, lm_input_len.to(device))
+        logits = self.llm_decoder(lm_output)
+        loss = self.criterion_ce(logits, lm_target)
+        acc = th_accuracy(logits.view(-1, self.speech_token_size + 1), lm_target, ignore_label=IGNORE_ID)
+        return {'loss': loss, 'acc': acc}
+
+    def sampling_ids(
+            self,
+            weighted_scores: torch.Tensor,
+            decoded_tokens: List,
+            sampling: int,
+            ignore_eos: bool = True,
+    ):
+        num_trials, max_trials = 0, 100
+        while True:
+            top_ids = self.sampling(weighted_scores, decoded_tokens, sampling)
+            if (not ignore_eos) or (self.speech_token_size not in top_ids):
+                break
+            num_trials += 1
+            if num_trials > max_trials:
+                raise RuntimeError('sampling reaches max_trials {} and still get eos when ignore_eos is True, check your input!'.format(max_trials))
+        return top_ids
+
+    @torch.inference_mode()
+    def inference(
+            self,
+            text: torch.Tensor,
+            text_len: torch.Tensor,
+            prompt_text: torch.Tensor,
+            prompt_text_len: torch.Tensor,
+            prompt_speech_token: torch.Tensor,
+            prompt_speech_token_len: torch.Tensor,
+            embedding: torch.Tensor,
+            sampling: int = 25,
+            max_token_text_ratio: float = 20,
+            min_token_text_ratio: float = 2,
+    ) -> Generator[torch.Tensor, None, None]:
+        if self.fp16 is True:
+            embedding = embedding.half()
+
+        device = text.device
+        text = torch.concat([prompt_text, text], dim=1)
+        text_len += prompt_text_len
+        text = self.text_embedding(text)
+
+        # 1. encode text
+        text, text_len = self.encode(text, text_len)
+
+        # 2. encode embedding
+        if embedding.shape[0] != 0:
+            embedding = F.normalize(embedding, dim=1)
+            embedding = self.spk_embed_affine_layer(embedding)
+            embedding = embedding.unsqueeze(dim=1)
+        else:
+            embedding = torch.zeros(1, 0, self.llm_input_size, dtype=text.dtype).to(device).to(text.dtype)
+
+        # 3. concat llm_input
+        sos_eos_emb = self.llm_embedding.weight[self.sos_eos].reshape(1, 1, -1)
+        task_id_emb = self.llm_embedding.weight[self.task_id].reshape(1, 1, -1)
+        if prompt_speech_token_len != 0:
+            prompt_speech_token_emb = self.speech_embedding(prompt_speech_token)
+        else:
+            prompt_speech_token_emb = torch.zeros(1, 0, self.llm_input_size, dtype=text.dtype).to(device)
+        lm_input = torch.concat([sos_eos_emb, embedding, text, task_id_emb, prompt_speech_token_emb], dim=1)
+
+        # 4. cal min/max_length
+        min_len = int((text_len - prompt_text_len) * min_token_text_ratio)
+        max_len = int((text_len - prompt_text_len) * max_token_text_ratio)
+
+        # 5. step by step decode
+        out_tokens = []
+        offset = 0
+        att_cache, cnn_cache = torch.zeros((0, 0, 0, 0), device=lm_input.device), torch.zeros((0, 0, 0, 0), device=lm_input.device)
+        for i in range(max_len):
+            y_pred, att_cache, cnn_cache = self.llm.forward_chunk(lm_input, offset=offset, required_cache_size=-1,
+                                                                  att_cache=att_cache, cnn_cache=cnn_cache,
+                                                                  att_mask=torch.tril(torch.ones((1, lm_input.shape[1], lm_input.shape[1]),
+                                                                                                 device=lm_input.device)).to(torch.bool))
+            logp = self.llm_decoder(y_pred[:, -1]).log_softmax(dim=-1)
+            # force continue decode first token
+            if i == 0:
+                logp[:, self.speech_token_size] = -float('inf')
+            top_ids = self.sampling_ids(logp.squeeze(dim=0), out_tokens, sampling, ignore_eos=True if i < min_len else False).item()
+            if top_ids == self.speech_token_size:
+                break
+            # in stream mode, yield token one by one
+            yield top_ids
+            out_tokens.append(top_ids)
+            offset += lm_input.size(1)
+            lm_input = self.speech_embedding.weight[top_ids].reshape(1, 1, -1)
+
+
+class Qwen2Encoder(torch.nn.Module):
+    def __init__(self, pretrain_path):
+        super().__init__()
+        self.model = Qwen2ForCausalLM.from_pretrained(pretrain_path)
+
+    def forward_one_step(self, xs, masks, cache=None):
+        input_masks = masks[:, -1, :]
+        outs = self.model(
+            inputs_embeds=xs,
+            attention_mask=input_masks,
+            output_hidden_states=True,
+            return_dict=True,
+            use_cache=True,
+            past_key_values=cache,
+        )
+        xs = outs.hidden_states[-1]
+        new_cache = outs.past_key_values
+        return xs, new_cache
+
+
+class Qwen2LM(TransformerLM):
+    def __init__(
+            self,
+            llm_input_size: int,
+            llm_output_size: int,
+            speech_token_size: int,
+            llm: torch.nn.Module,
+            sampling: Callable,
+            length_normalized_loss: bool = True,
+            lsm_weight: float = 0.0,
+            mix_ratio: List[int] = [5, 15],
+            dpo: bool = False,
+    ):
+        torch.nn.Module.__init__(self)
+        self.llm_input_size = llm_input_size
+        self.llm_output_size = llm_output_size
+        self.speech_token_size = speech_token_size
+
+        # 2. build speech token language model related modules
+        self.sos_eos = 0
+        self.task_id = 1
+        self.fill_token = 2
+
+        self.llm_embedding = torch.nn.Embedding(2, llm_input_size)
+        self.llm = llm
+        self.llm_decoder = nn.Linear(llm_output_size, speech_token_size + 3)
+        self.criterion_ce = LabelSmoothingLoss(
+            size=speech_token_size + 3,
+            padding_idx=IGNORE_ID,
+            smoothing=lsm_weight,
+            normalize_length=length_normalized_loss,
+        )
+
+        # 3. [Optional] build speech token related modules
+        self.speech_embedding = torch.nn.Embedding(speech_token_size + 3, llm_input_size)
+
+        # 4. sampling method
+        self.sampling = sampling
+        self.mix_ratio = mix_ratio
+
+        # 5. [Optional] set dpo
+        self.dpo = dpo
+
+
+    def forward(
+            self,
+            batch: dict,
+            device: torch.device,
+        ) -> Dict[str, Optional[torch.Tensor]]:
+        text_token = batch['text_token'].to(device)
+        text_token_len = batch['text_token_len'].to(device)
+        speech_token = batch['speech_token'].to(device)
+        speech_token_len = batch['speech_token_len'].to(device)
+        if self.dpo:
+            reject_speech_token = batch['reject_speech_token'].to(device)
+            reject_speech_token_len = batch['reject_speech_token_len'].to(device)
+        # 1. prepare llm_target
+        sos_eos_emb = self.llm_embedding.weight[self.sos_eos].reshape(1, 1, -1)
+        task_id_emb = self.llm_embedding.weight[self.task_id].reshape(1, 1, -1)
+        target_ids = [torch.tensor([IGNORE_ID] * (1 + text_token_len[i]) + speech_token[i, :speech_token_len[i]].tolist() +
+                                        [self.speech_token_size]) for i in range(text_token.size(0))]
+        if self.dpo:
+            reject_target_ids = [torch.tensor([IGNORE_ID] * (1 + text_token_len[i]) + reject_speech_token[i, :reject_speech_token_len[i]].tolist() +
+                                            [self.speech_token_size]) for i in range(text_token.size(0))]
+            target_ids.extend(reject_target_ids)
+        target_ids = pad_sequence(target_ids, batch_first=True, padding_value=IGNORE_ID).to(device)
+
+        # 2. speech token projection
+        speech_emb = self.speech_embedding(speech_token)
+        if self.dpo:
+            reject_speech_emb = self.speech_embedding(reject_speech_token)
+
+        # 3. text token projection
+        text_token_lst = unpad_sequence(text_token, text_token_len, batch_first=True)
+        text_emb = [self.llm.model.model.embed_tokens(y) for y in text_token_lst]
+
+        # 4. prepare llm_input
+        speech_emb = unpad_sequence(speech_emb, speech_token_len.cpu(), batch_first=True)
+        input_emb = [torch.concat([sos_eos_emb.squeeze(dim=0), text_emb[i], task_id_emb.squeeze(dim=0), speech_emb[i]], dim=0)
+                     for i in range(len(text_emb))]
+        if self.dpo:
+            reject_speech_emb = unpad_sequence(reject_speech_emb, reject_speech_token_len.cpu(), batch_first=True)
+            reject_input_emb = [torch.concat([sos_eos_emb.squeeze(dim=0), text_emb[i], task_id_emb.squeeze(dim=0), reject_speech_emb[i]], dim=0)
+                                for i in range(len(text_emb))]
+            input_emb.extend(reject_input_emb)
+        input_emb_lengths = torch.tensor([i.size(0) for i in input_emb], dtype=torch.int32).to(device)
+        input_emb = pad_sequence(input_emb, batch_first=True, padding_value=IGNORE_ID).to(device)
+
+        attention_mask = ~make_pad_mask(input_emb_lengths)
+
+        result = self.llm.model(
+            inputs_embeds=input_emb,
+            attention_mask=attention_mask,
+            return_dict=True
+        )
+        hidden_states = result.hidden_states
+        logits = self.llm_decoder(hidden_states[-1])
+        loss = self.criterion_ce(logits[: speech_token.shape[0]], target_ids[: speech_token.shape[0]])
+        acc = th_accuracy(
+            logits[: speech_token.shape[0]].view(-1, self.speech_token_size + 3),
+            target_ids[: speech_token.shape[0]],
+            ignore_label=IGNORE_ID,
+        )
+        if not self.dpo:
+            return {
+                "loss": loss,
+                "acc": acc,
+            }
+        else:
+            all_logps_sum, all_logps_mean = self.get_batch_logps(
+                logits, target_ids, attention_mask, text_token_len, average_log_prob=False, ignore_id=IGNORE_ID
+            )
+            chosen_logps = all_logps_sum[: speech_token.shape[0]]
+            rejected_logps = all_logps_sum[speech_token.shape[0]:]
+            return {
+                "loss": loss,
+                "acc": acc,
+                "chosen_logps": chosen_logps,
+                "rejected_logps": rejected_logps
+            }
+
+
+    def get_batch_logps(
+        self,
+        logits: torch.FloatTensor,
+        labels: torch.LongTensor,
+        attention_mask,
+        prompt_token_lens,
+        average_log_prob: bool = False,
+        ignore_id: int = -1,
+    ) -> torch.FloatTensor:
+        """Compute the log probabilities of the given labels under the given logits.
+
+        Args:
+            logits: Logits of the model (unnormalized). Shape: (batch_size, sequence_length, vocab_size)
+            labels: Labels for which to compute the log probabilities. Label tokens with a value of -100 are ignored. Shape: (batch_size, sequence_length)
+            average_log_prob: If True, return the average log probability per (non-masked) token. Otherwise, return the sum of the log probabilities of the (non-masked) tokens.
+
+        Returns:
+            A tensor of shape (batch_size,) containing the average/sum log probabilities of the given labels under the given logits.
+        """
+        assert average_log_prob == False
+        assert logits.shape[:-1] == labels.shape
+        labels = labels[:, 1:].clone()
+        logits = logits[:, :-1, :]
+        loss_masks = attention_mask.clone().bool()
+        # mask prompts
+        for mask, text_token_len in zip(loss_masks, prompt_token_lens):
+            mask[:text_token_len + 1] = False
+        loss_masks = loss_masks[:, 1:]
+        labels[loss_masks == False] = 0
+        # dummy token; we'll ignore the losses on these tokens later
+        ignore = labels == ignore_id
+        labels = labels.masked_fill(ignore, 0)  # avoid -1 index
+        per_token_logps = torch.gather(logits.log_softmax(-1), dim=2, index=labels.unsqueeze(2)).squeeze(2)   # (bs, time,)
+        logprobs_sums = (per_token_logps * loss_masks).sum(-1)
+        logprobs_means = (per_token_logps * loss_masks).sum(-1) / loss_masks.sum(-1)
+        return logprobs_sums, logprobs_means
+
+
+    @torch.inference_mode()
+    def inference(
+            self,
+            text: torch.Tensor,
+            text_len: torch.Tensor,
+            prompt_text: torch.Tensor,
+            prompt_text_len: torch.Tensor,
+            prompt_speech_token: torch.Tensor,
+            prompt_speech_token_len: torch.Tensor,
+            embedding: torch.Tensor,
+            sampling: int = 25,
+            max_token_text_ratio: float = 20,
+            min_token_text_ratio: float = 2,
+    ) -> Generator[torch.Tensor, None, None]:
+        device = text.device
+        text = torch.concat([prompt_text, text], dim=1)
+        text_len += prompt_text_len
+        text = self.llm.model.model.embed_tokens(text)
+
+        # 3. concat llm_input
+        sos_eos_emb = self.llm_embedding.weight[self.sos_eos].reshape(1, 1, -1)
+        task_id_emb = self.llm_embedding.weight[self.task_id].reshape(1, 1, -1)
+        if prompt_speech_token_len != 0:
+            prompt_speech_token_emb = self.speech_embedding(prompt_speech_token)
+        else:
+            prompt_speech_token_emb = torch.zeros(1, 0, self.llm_input_size, dtype=text.dtype).to(device)
+        lm_input = torch.concat([sos_eos_emb, text, task_id_emb, prompt_speech_token_emb], dim=1)
+
+        # 4. cal min/max_length
+        min_len = int((text_len - prompt_text_len) * min_token_text_ratio)
+        max_len = int((text_len - prompt_text_len) * max_token_text_ratio)
+
+        # 5. step by step decode
+        out_tokens = []
+        cache = None
+        for i in range(max_len):
+            y_pred, cache = self.llm.forward_one_step(lm_input,
+                                                      masks=torch.tril(torch.ones((1, lm_input.shape[1], lm_input.shape[1]), device=lm_input.device)).to(torch.bool),
+                                                      cache=cache)
+            logp = self.llm_decoder(y_pred[:, -1]).log_softmax(dim=-1)
+            top_ids = self.sampling_ids(logp.squeeze(dim=0), out_tokens, sampling, ignore_eos=True if i < min_len else False).item()
+            if top_ids == self.speech_token_size:
+                break
+            if top_ids > self.speech_token_size:
+                continue
+            # in stream mode, yield token one by one
+            yield top_ids
+            out_tokens.append(top_ids)
+            lm_input = self.speech_embedding.weight[top_ids].reshape(1, 1, -1)
+
+    @torch.inference_mode()
+    def inference_bistream(
+            self,
+            text: Generator,
+            prompt_text: torch.Tensor,
+            prompt_text_len: torch.Tensor,
+            prompt_speech_token: torch.Tensor,
+            prompt_speech_token_len: torch.Tensor,
+            embedding: torch.Tensor,
+            sampling: int = 25,
+            max_token_text_ratio: float = 20,
+            min_token_text_ratio: float = 2,
+    ) -> Generator[torch.Tensor, None, None]:
+
+        device = prompt_text.device
+        # 1. prepare input
+        sos_eos_emb = self.llm_embedding.weight[self.sos_eos].reshape(1, 1, -1)
+        task_id_emb = self.llm_embedding.weight[self.task_id].reshape(1, 1, -1)
+        if prompt_speech_token_len != 0:
+            prompt_speech_token_emb = self.speech_embedding(prompt_speech_token)
+        else:
+            prompt_speech_token_emb = torch.zeros(1, 0, self.llm_input_size, dtype=prompt_text.dtype).to(device)
+        lm_input = torch.concat([sos_eos_emb], dim=1)
+
+        # 2. iterate text
+        out_tokens = []
+        cache = None
+        # NOTE init prompt_text as text_cache as it is basically impossible prompt_speech_token/prompt_text < 15/5
+        text_cache = self.llm.model.model.embed_tokens(prompt_text)
+        next_fill_index = -1
+        for this_text in text:
+            text_cache = torch.concat([text_cache, self.llm.model.model.embed_tokens(this_text)], dim=1)
+            # prompt_speech_token_emb not empty, try append to lm_input
+            while prompt_speech_token_emb.size(1) != 0:
+                if text_cache.size(1) >= self.mix_ratio[0]:
+                    lm_input_text, lm_input_speech = text_cache[:, :self.mix_ratio[0]], prompt_speech_token_emb[:, :self.mix_ratio[1]]
+                    logging.info('append {} text token {} speech token'.format(lm_input_text.size(1), lm_input_speech.size(1)))
+                    lm_input = torch.concat([lm_input, lm_input_text, lm_input_speech], dim=1)
+                    text_cache, prompt_speech_token_emb = text_cache[:, self.mix_ratio[0]:], prompt_speech_token_emb[:, self.mix_ratio[1]:]
+                else:
+                    logging.info('not enough text token to decode, wait for more')
+                    break
+            # no prompt_speech_token_emb remain, can decode some speech token
+            if prompt_speech_token_emb.size(1) == 0:
+                if (len(out_tokens) != 0 and out_tokens[-1] == self.speech_token_size + 2) or (len(out_tokens) == 0 and lm_input.size(1) == 1):
+                    logging.info('get fill token, need to append more text token')
+                    if text_cache.size(1) >= self.mix_ratio[0]:
+                        lm_input_text = text_cache[:, :self.mix_ratio[0]]
+                        logging.info('append {} text token'.format(lm_input_text.size(1)))
+                        if len(out_tokens) != 0 and out_tokens[-1] == self.speech_token_size + 2:
+                            lm_input = lm_input_text
+                        else:
+                            lm_input = torch.concat([lm_input, lm_input_text], dim=1)
+                        text_cache = text_cache[:, self.mix_ratio[0]:]
+                    else:
+                        logging.info('not enough text token to decode, wait for more')
+                        continue
+                while True:
+                    seq_len = lm_input.shape[1] if cache is None else lm_input.shape[1] + cache[0][0].size(2)
+                    y_pred, cache = self.llm.forward_one_step(lm_input,
+                                                masks=torch.tril(torch.ones((1, seq_len, seq_len), device=lm_input.device)).to(torch.bool),
+                                                cache=cache)
+                    logp = self.llm_decoder(y_pred[:, -1]).log_softmax(dim=-1)
+                    if next_fill_index != -1 and len(out_tokens) == next_fill_index:
+                        top_ids = self.speech_token_size + 2
+                        next_fill_index += (self.mix_ratio[1] + 1)
+                    else:
+                        top_ids = self.sampling_ids(logp.squeeze(dim=0), out_tokens, sampling, ignore_eos=True).item()
+                    if top_ids == self.speech_token_size + 2:
+                        next_fill_index = len(out_tokens) + self.mix_ratio[1] + 1
+                        logging.info('fill_token index {} next fill_token index {}'.format(len(out_tokens), next_fill_index))
+                    out_tokens.append(top_ids)
+                    if top_ids >= self.speech_token_size:
+                        if top_ids == self.speech_token_size + 2:
+                            break
+                        else:
+                            raise ValueError('should not get token {}'.format(top_ids))
+                    yield top_ids
+                    lm_input = self.speech_embedding.weight[top_ids].reshape(1, 1, -1)
+
+        # 3. final decode
+        lm_input = torch.concat([lm_input, text_cache, task_id_emb], dim=1)
+        logging.info('no more text token, decode until met eos')
+        while True:
+            seq_len = lm_input.shape[1] if cache is None else lm_input.shape[1] + cache[0][0].size(2)
+            y_pred, cache = self.llm.forward_one_step(lm_input,
+                                                      masks=torch.tril(torch.ones((1, seq_len, seq_len), device=lm_input.device)).to(torch.bool),
+                                                      cache=cache)
+            logp = self.llm_decoder(y_pred[:, -1]).log_softmax(dim=-1)
+            top_ids = self.sampling_ids(logp.squeeze(dim=0), out_tokens, sampling, ignore_eos=False).item()
+            out_tokens.append(top_ids)
+            if top_ids >= self.speech_token_size:
+                if top_ids == self.speech_token_size:
+                    break
+                else:
+                    raise ValueError('should not get token {}'.format(top_ids))
+            # in stream mode, yield token one by one
+            yield top_ids
+            lm_input = self.speech_embedding.weight[top_ids].reshape(1, 1, -1)

cosyvoice/utils/executor_dpo.py

@@ -0,0 +1,184 @@
+# Copyright (c) 2020 Mobvoi Inc (Binbin Zhang)
+#               2024 Alibaba Inc (authors: Xiang Lyu)
+#
+# 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 logging
+from contextlib import nullcontext
+import os
+
+import torch
+import torch.distributed as dist
+
+from cosyvoice.utils.train_utils_dpo import update_parameter_and_lr, log_per_step, log_per_save, batch_forward, batch_backward, save_model, cosyvoice_join
+from cosyvoice.utils.losses_dpo import DPOLoss
+
+
+class Executor:
+
+    def __init__(self, gan: bool = False, dpo: bool = False, beta: float = 0.01, label_smoothing: float = 0.0, ipo: bool = False):
+        self.gan = gan
+        self.step = 0
+        self.epoch = 0
+        self.rank = int(os.environ.get('RANK', 0))
+        self.device = torch.device('cuda:{}'.format(self.rank))
+        self.dpo = dpo
+        if self.dpo:
+            self.dpo_loss = DPOLoss(beta, label_smoothing, ipo)
+        else:
+            self.dpo_loss = None
+
+    def train_one_epoc(self, model, optimizer, scheduler, train_data_loader, cv_data_loader, writer, info_dict, scaler, group_join, ref_model=None):
+        ''' Train one epoch
+        '''
+
+        lr = optimizer.param_groups[0]['lr']
+        logging.info('Epoch {} TRAIN info lr {} rank {}'.format(self.epoch, lr, self.rank))
+        logging.info('using accumulate grad, new batch size is {} times'
+                     ' larger than before'.format(info_dict['accum_grad']))
+        # A context manager to be used in conjunction with an instance of
+        # torch.nn.parallel.DistributedDataParallel to be able to train
+        # with uneven inputs across participating processes.
+        model.train()
+        if self.dpo:
+            assert ref_model is not None
+            ref_model.eval()
+        model_context = model.join if info_dict['train_engine'] == 'torch_ddp' else nullcontext
+        with model_context():
+            for batch_idx, batch_dict in enumerate(train_data_loader):
+                info_dict["tag"] = "TRAIN"
+                info_dict["step"] = self.step
+                info_dict["epoch"] = self.epoch
+                info_dict["batch_idx"] = batch_idx
+                if cosyvoice_join(group_join, info_dict):
+                    break
+
+                # Disable gradient synchronizations across DDP processes.
+                # Within this context, gradients will be accumulated on module
+                # variables, which will later be synchronized.
+                if info_dict['train_engine'] == 'torch_ddp' and (batch_idx + 1) % info_dict["accum_grad"] != 0:
+                    context = model.no_sync
+                # Used for single gpu training and DDP gradient synchronization
+                # processes.
+                else:
+                    context = nullcontext
+
+                with context():
+                    info_dict = batch_forward(model, batch_dict, scaler, info_dict, ref_model, self.dpo_loss)
+                    info_dict = batch_backward(model, scaler, info_dict)
+
+                info_dict = update_parameter_and_lr(model, optimizer, scheduler, scaler, info_dict)
+                log_per_step(writer, info_dict)
+                # NOTE specify save_per_step in cosyvoice.yaml if you want to enable step save
+                if info_dict['save_per_step'] > 0 and (self.step + 1) % info_dict['save_per_step'] == 0 and \
+                   (batch_idx + 1) % info_dict["accum_grad"] == 0:
+                    dist.barrier()
+                    self.cv(model, cv_data_loader, writer, info_dict, on_batch_end=False, ref_model=ref_model, dpo_loss=self.dpo_loss)
+                    model.train()
+                if (batch_idx + 1) % info_dict["accum_grad"] == 0:
+                    self.step += 1
+        dist.barrier()
+        self.cv(model, cv_data_loader, writer, info_dict, on_batch_end=True, ref_model=ref_model, dpo_loss=self.dpo_loss)
+
+    def train_one_epoc_gan(self, model, optimizer, scheduler, optimizer_d, scheduler_d, train_data_loader, cv_data_loader,
+                           writer, info_dict, scaler, group_join):
+        ''' Train one epoch
+        '''
+
+        lr = optimizer.param_groups[0]['lr']
+        logging.info('Epoch {} TRAIN info lr {} rank {}'.format(self.epoch, lr, self.rank))
+        logging.info('using accumulate grad, new batch size is {} times'
+                     ' larger than before'.format(info_dict['accum_grad']))
+        # A context manager to be used in conjunction with an instance of
+        # torch.nn.parallel.DistributedDataParallel to be able to train
+        # with uneven inputs across participating processes.
+        model.train()
+        model_context = model.join if info_dict['train_engine'] == 'torch_ddp' else nullcontext
+        with model_context():
+            for batch_idx, batch_dict in enumerate(train_data_loader):
+                info_dict["tag"] = "TRAIN"
+                info_dict["step"] = self.step
+                info_dict["epoch"] = self.epoch
+                info_dict["batch_idx"] = batch_idx
+                if cosyvoice_join(group_join, info_dict):
+                    break
+
+                # Disable gradient synchronizations across DDP processes.
+                # Within this context, gradients will be accumulated on module
+                # variables, which will later be synchronized.
+                if info_dict['train_engine'] == 'torch_ddp' and (batch_idx + 1) % info_dict["accum_grad"] != 0:
+                    context = model.no_sync
+                # Used for single gpu training and DDP gradient synchronization
+                # processes.
+                else:
+                    context = nullcontext
+
+                with context():
+                    batch_dict['turn'] = 'discriminator'
+                    info_dict = batch_forward(model, batch_dict, scaler, info_dict)
+                    info_dict = batch_backward(model, scaler, info_dict)
+                info_dict = update_parameter_and_lr(model, optimizer_d, scheduler_d, scaler, info_dict)
+                optimizer.zero_grad()
+                log_per_step(writer, info_dict)
+                with context():
+                    batch_dict['turn'] = 'generator'
+                    info_dict = batch_forward(model, batch_dict, scaler, info_dict)
+                    info_dict = batch_backward(model, scaler, info_dict)
+                info_dict = update_parameter_and_lr(model, optimizer, scheduler, scaler, info_dict)
+                optimizer_d.zero_grad()
+                log_per_step(writer, info_dict)
+                # NOTE specify save_per_step in cosyvoice.yaml if you want to enable step save
+                if info_dict['save_per_step'] > 0 and (self.step + 1) % info_dict['save_per_step'] == 0 and \
+                   (batch_idx + 1) % info_dict["accum_grad"] == 0:
+                    dist.barrier()
+                    self.cv(model, cv_data_loader, writer, info_dict, on_batch_end=False)
+                    model.train()
+                if (batch_idx + 1) % info_dict["accum_grad"] == 0:
+                    self.step += 1
+        dist.barrier()
+        self.cv(model, cv_data_loader, writer, info_dict, on_batch_end=True)
+
+    @torch.inference_mode()
+    def cv(self, model, cv_data_loader, writer, info_dict, on_batch_end=True, ref_model=None, dpo_loss=None):
+        ''' Cross validation on
+        '''
+        logging.info('Epoch {} Step {} on_batch_end {} CV rank {}'.format(self.epoch, self.step + 1, on_batch_end, self.rank))
+        model.eval()
+        if self.dpo:
+            assert ref_model is not None
+            ref_model.eval()
+        total_num_utts, total_loss_dict = 0, {}  # avoid division by 0
+        for batch_idx, batch_dict in enumerate(cv_data_loader):
+            info_dict["tag"] = "CV"
+            info_dict["step"] = self.step
+            info_dict["epoch"] = self.epoch
+            info_dict["batch_idx"] = batch_idx
+
+            num_utts = len(batch_dict["utts"])
+            total_num_utts += num_utts
+
+            if self.gan is True:
+                batch_dict['turn'] = 'generator'
+            info_dict = batch_forward(model, batch_dict, None, info_dict, ref_model, dpo_loss)
+
+            for k, v in info_dict['loss_dict'].items():
+                if k not in total_loss_dict:
+                    total_loss_dict[k] = []
+                total_loss_dict[k].append(v.item() * num_utts)
+            log_per_step(None, info_dict)
+        for k, v in total_loss_dict.items():
+            total_loss_dict[k] = sum(v) / total_num_utts
+        info_dict['loss_dict'] = total_loss_dict
+        log_per_save(writer, info_dict)
+        model_name = 'epoch_{}_whole'.format(self.epoch) if on_batch_end else 'epoch_{}_step_{}'.format(self.epoch, self.step + 1)
+        save_model(model, model_name, info_dict)

cosyvoice/utils/losses_dpo.py

@@ -0,0 +1,57 @@
+import torch
+import torch.nn.functional as F
+from typing import Tuple
+
+
+def tpr_loss(disc_real_outputs, disc_generated_outputs, tau):
+    loss = 0
+    for dr, dg in zip(disc_real_outputs, disc_generated_outputs):
+        m_DG = torch.median((dr - dg))
+        L_rel = torch.mean((((dr - dg) - m_DG) ** 2)[dr < dg + m_DG])
+        loss += tau - F.relu(tau - L_rel)
+    return loss
+
+
+def mel_loss(real_speech, generated_speech, mel_transforms):
+    loss = 0
+    for transform in mel_transforms:
+        mel_r = transform(real_speech)
+        mel_g = transform(generated_speech)
+        loss += F.l1_loss(mel_g, mel_r)
+    return loss
+
+
+class DPOLoss(torch.nn.Module):
+    """
+    DPO Loss
+    """
+
+    def __init__(self, beta: float, label_smoothing: float = 0.0, ipo: bool = False) -> None:
+        super().__init__()
+        self.beta = beta
+        self.label_smoothing = label_smoothing
+        self.ipo = ipo
+
+    def forward(
+        self,
+        policy_chosen_logps: torch.Tensor,
+        policy_rejected_logps: torch.Tensor,
+        reference_chosen_logps: torch.Tensor,
+        reference_rejected_logps: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        pi_logratios = policy_chosen_logps - policy_rejected_logps
+        ref_logratios = reference_chosen_logps - reference_rejected_logps
+        logits = pi_logratios - ref_logratios
+        if self.ipo:
+            losses = (logits - 1 / (2 * self.beta)) ** 2  # Eq. 17 of https://arxiv.org/pdf/2310.12036v2.pdf
+        else:
+            # Eq. 3 https://ericmitchell.ai/cdpo.pdf; label_smoothing=0 gives original DPO (Eq. 7 of https://arxiv.org/pdf/2305.18290.pdf)
+            losses = (
+                -F.logsigmoid(self.beta * logits) * (1 - self.label_smoothing)
+                - F.logsigmoid(-self.beta * logits) * self.label_smoothing
+            )
+        loss = losses.mean()
+        chosen_rewards = self.beta * (policy_chosen_logps - reference_chosen_logps).detach()
+        rejected_rewards = self.beta * (policy_rejected_logps - reference_rejected_logps).detach()
+
+        return loss, chosen_rewards, rejected_rewards

cosyvoice/utils/train_utils_dpo.py

@@ -0,0 +1,364 @@
+# Copyright (c) 2021 Mobvoi Inc. (authors: Binbin Zhang)
+#               2023 Horizon Inc. (authors: Xingchen Song)
+#               2024 Alibaba Inc (authors: Xiang Lyu)
+#
+# 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 logging
+import os
+import torch
+import json
+import re
+import datetime
+import yaml
+
+import deepspeed
+import torch.optim as optim
+import torch.distributed as dist
+
+from torch.utils.tensorboard import SummaryWriter
+from torch.utils.data import DataLoader
+from torch.nn.utils import clip_grad_norm_
+
+from deepspeed.runtime.zero.stage_1_and_2 import estimate_zero2_model_states_mem_needs_all_live
+
+from cosyvoice.dataset.dataset import Dataset
+from cosyvoice.utils.scheduler import WarmupLR, NoamHoldAnnealing, ConstantLR
+
+
+def init_distributed(args):
+    world_size = int(os.environ.get('WORLD_SIZE', 1))
+    local_rank = int(os.environ.get('LOCAL_RANK', 0))
+    rank = int(os.environ.get('RANK', 0))
+    logging.info('training on multiple gpus, this gpu {}'.format(local_rank) +
+                 ', rank {}, world_size {}'.format(rank, world_size))
+    if args.train_engine == 'torch_ddp':
+        torch.cuda.set_device(local_rank)
+        dist.init_process_group(args.dist_backend)
+    else:
+        deepspeed.init_distributed(dist_backend=args.dist_backend)
+    return world_size, local_rank, rank
+
+
+def init_dataset_and_dataloader(args, configs, gan):
+    data_pipeline = configs['data_pipeline_gan'] if gan is True else configs['data_pipeline']
+    train_dataset = Dataset(args.train_data, data_pipeline=data_pipeline, mode='train', gan=gan, shuffle=True, partition=True)
+    cv_dataset = Dataset(args.cv_data, data_pipeline=data_pipeline, mode='train', gan=gan, shuffle=False, partition=False)
+
+    # do not use persistent_workers=True, as whisper tokenizer opens tiktoken file each time when the for loop starts
+    train_data_loader = DataLoader(train_dataset,
+                                   batch_size=None,
+                                   pin_memory=args.pin_memory,
+                                   num_workers=args.num_workers,
+                                   prefetch_factor=args.prefetch)
+    cv_data_loader = DataLoader(cv_dataset,
+                                batch_size=None,
+                                pin_memory=args.pin_memory,
+                                num_workers=args.num_workers,
+                                prefetch_factor=args.prefetch)
+    return train_dataset, cv_dataset, train_data_loader, cv_data_loader
+
+
+def check_modify_and_save_config(args, configs):
+    if args.train_engine == "torch_ddp":
+        configs['train_conf']["dtype"] = 'fp32'
+    else:
+        with open(args.deepspeed_config, 'r') as fin:
+            ds_configs = json.load(fin)
+        if "fp16" in ds_configs and ds_configs["fp16"]["enabled"]:
+            configs['train_conf']["dtype"] = "fp16"
+        elif "bf16" in ds_configs and ds_configs["bf16"]["enabled"]:
+            configs['train_conf']["dtype"] = "bf16"
+        else:
+            configs['train_conf']["dtype"] = "fp32"
+        assert ds_configs["train_micro_batch_size_per_gpu"] == 1
+        # if use deepspeed, override ddp config
+        configs['train_conf']['save_per_step'] = int(configs['train_conf']['save_per_step'] *
+                                                     configs['train_conf']['accum_grad'] / ds_configs["gradient_accumulation_steps"])
+        configs['train_conf']['accum_grad'] = ds_configs["gradient_accumulation_steps"]
+        configs['train_conf']['grad_clip'] = ds_configs["gradient_clipping"]
+        configs['train_conf']['log_interval'] = ds_configs["steps_per_print"]
+    return configs
+
+
+def wrap_cuda_model(args, model):
+    local_world_size = int(os.environ.get('LOCAL_WORLD_SIZE', 1))
+    world_size = int(os.environ.get('WORLD_SIZE', 1))
+    if args.train_engine == "torch_ddp":  # native pytorch ddp
+        assert (torch.cuda.is_available())
+        model.cuda()
+        model = torch.nn.parallel.DistributedDataParallel(model, find_unused_parameters=True)
+    else:
+        if int(os.environ.get('RANK', 0)) == 0:
+            logging.info("Estimating model states memory needs (zero2)...")
+            estimate_zero2_model_states_mem_needs_all_live(
+                model,
+                num_gpus_per_node=local_world_size,
+                num_nodes=world_size // local_world_size)
+    return model
+
+
+def init_optimizer_and_scheduler(args, configs, model, gan):
+    if gan is False:
+        if configs['train_conf']['optim'] == 'adam':
+            optimizer = optim.Adam(model.parameters(), **configs['train_conf']['optim_conf'])
+        elif configs['train_conf']['optim'] == 'adamw':
+            optimizer = optim.AdamW(model.parameters(), **configs['train_conf']['optim_conf'])
+        else:
+            raise ValueError("unknown optimizer: " + configs['train_conf'])
+
+        if configs['train_conf']['scheduler'] == 'warmuplr':
+            scheduler_type = WarmupLR
+            scheduler = WarmupLR(optimizer, **configs['train_conf']['scheduler_conf'])
+        elif configs['train_conf']['scheduler'] == 'NoamHoldAnnealing':
+            scheduler_type = NoamHoldAnnealing
+            scheduler = NoamHoldAnnealing(optimizer, **configs['train_conf']['scheduler_conf'])
+        elif configs['train_conf']['scheduler'] == 'constantlr':
+            scheduler_type = ConstantLR
+            scheduler = ConstantLR(optimizer)
+        else:
+            raise ValueError("unknown scheduler: " + configs['train_conf'])
+
+        # use deepspeed optimizer for speedup
+        if args.train_engine == "deepspeed":
+            def scheduler(opt):
+                return scheduler_type(opt, **configs['train_conf']['scheduler_conf'])
+            model, optimizer, _, scheduler = deepspeed.initialize(
+                args=args,
+                model=model,
+                optimizer=None,
+                lr_scheduler=scheduler,
+                model_parameters=model.parameters())
+
+        optimizer_d, scheduler_d = None, None
+
+    else:
+        # currently we wrap generator and discriminator in one model, so we cannot use deepspeed
+        if configs['train_conf']['optim'] == 'adam':
+            optimizer = optim.Adam(model.module.generator.parameters(), **configs['train_conf']['optim_conf'])
+        elif configs['train_conf']['optim'] == 'adamw':
+            optimizer = optim.AdamW(model.module.generator.parameters(), **configs['train_conf']['optim_conf'])
+        else:
+            raise ValueError("unknown optimizer: " + configs['train_conf'])
+
+        if configs['train_conf']['scheduler'] == 'warmuplr':
+            scheduler_type = WarmupLR
+            scheduler = WarmupLR(optimizer, **configs['train_conf']['scheduler_conf'])
+        elif configs['train_conf']['scheduler'] == 'NoamHoldAnnealing':
+            scheduler_type = NoamHoldAnnealing
+            scheduler = NoamHoldAnnealing(optimizer, **configs['train_conf']['scheduler_conf'])
+        elif configs['train_conf']['scheduler'] == 'constantlr':
+            scheduler_type = ConstantLR
+            scheduler = ConstantLR(optimizer)
+        else:
+            raise ValueError("unknown scheduler: " + configs['train_conf'])
+
+        if configs['train_conf']['optim_d'] == 'adam':
+            optimizer_d = optim.Adam(model.module.discriminator.parameters(), **configs['train_conf']['optim_conf'])
+        elif configs['train_conf']['optim_d'] == 'adamw':
+            optimizer_d = optim.AdamW(model.module.discriminator.parameters(), **configs['train_conf']['optim_conf'])
+        else:
+            raise ValueError("unknown optimizer: " + configs['train_conf'])
+
+        if configs['train_conf']['scheduler_d'] == 'warmuplr':
+            scheduler_type = WarmupLR
+            scheduler_d = WarmupLR(optimizer_d, **configs['train_conf']['scheduler_conf'])
+        elif configs['train_conf']['scheduler_d'] == 'NoamHoldAnnealing':
+            scheduler_type = NoamHoldAnnealing
+            scheduler_d = NoamHoldAnnealing(optimizer_d, **configs['train_conf']['scheduler_conf'])
+        elif configs['train_conf']['scheduler'] == 'constantlr':
+            scheduler_type = ConstantLR
+            scheduler_d = ConstantLR(optimizer_d)
+        else:
+            raise ValueError("unknown scheduler: " + configs['train_conf'])
+    return model, optimizer, scheduler, optimizer_d, scheduler_d
+
+
+def init_summarywriter(args):
+    writer = None
+    if int(os.environ.get('RANK', 0)) == 0:
+        os.makedirs(args.model_dir, exist_ok=True)
+        writer = SummaryWriter(args.tensorboard_dir)
+    return writer
+
+
+def save_model(model, model_name, info_dict):
+    rank = int(os.environ.get('RANK', 0))
+    model_dir = info_dict["model_dir"]
+    save_model_path = os.path.join(model_dir, '{}.pt'.format(model_name))
+
+    if info_dict["train_engine"] == "torch_ddp":
+        if rank == 0:
+            torch.save({**model.module.state_dict(), 'epoch': info_dict['epoch'], 'step': info_dict['step']}, save_model_path)
+    else:
+        with torch.no_grad():
+            model.save_checkpoint(save_dir=model_dir,
+                                  tag=model_name,
+                                  client_state=info_dict)
+    if rank == 0:
+        info_path = re.sub('.pt$', '.yaml', save_model_path)
+        info_dict['save_time'] = datetime.datetime.now().strftime('%d/%m/%Y %H:%M:%S')
+        with open(info_path, 'w') as fout:
+            data = yaml.dump(info_dict)
+            fout.write(data)
+        logging.info('[Rank {}] Checkpoint: save to checkpoint {}'.format(rank, save_model_path))
+
+
+def cosyvoice_join(group_join, info_dict):
+    world_size = int(os.environ.get('WORLD_SIZE', 1))
+    local_rank = int(os.environ.get('LOCAL_RANK', 0))
+    rank = int(os.environ.get('RANK', 0))
+
+    if info_dict["batch_idx"] != 0:
+        # we try to join all rank in both ddp and deepspeed mode, in case different rank has different lr
+        try:
+            dist.monitored_barrier(group=group_join,
+                                   timeout=group_join.options._timeout)
+            return False
+        except RuntimeError as e:
+            logging.info("Detected uneven workload distribution: {}\n".format(e) +
+                         "Break current worker to manually join all workers, " +
+                         "world_size {}, current rank {}, current local_rank {}\n".
+                         format(world_size, rank, local_rank))
+            return True
+    else:
+        return False
+
+
+def batch_forward(model, batch, scaler, info_dict, ref_model=None, dpo_loss=None):
+    device = int(os.environ.get('LOCAL_RANK', 0))
+
+    dtype = info_dict["dtype"]
+    if dtype == "fp16":
+        dtype = torch.float16
+    elif dtype == "bf16":
+        dtype = torch.bfloat16
+    else:  # fp32
+        dtype = torch.float32
+
+    if info_dict['train_engine'] == 'torch_ddp':
+        autocast = torch.cuda.amp.autocast(enabled=scaler is not None)
+    else:
+        autocast = torch.cuda.amp.autocast(enabled=True, dtype=dtype, cache_enabled=False)
+
+    with autocast:
+        info_dict['loss_dict'] = model(batch, device)
+        if ref_model and dpo_loss:
+            chosen_logps = info_dict['loss_dict']["chosen_logps"]
+            rejected_logps = info_dict['loss_dict']["rejected_logps"]
+            sft_loss = info_dict['loss_dict']['loss']
+            with torch.no_grad():
+                ref_model = ref_model.to(device)
+                ref_loss_dict = ref_model(batch, device)
+            reference_chosen_logps = ref_loss_dict["chosen_logps"]
+            reference_rejected_logps = ref_loss_dict["rejected_logps"]
+            preference_loss, chosen_reward, reject_reward = dpo_loss(
+                chosen_logps, rejected_logps, reference_chosen_logps, reference_rejected_logps
+            )
+            dpo_acc = (chosen_reward > reject_reward).float().mean()
+            info_dict['loss_dict']["loss"] = preference_loss + sft_loss
+            info_dict['loss_dict']["sft_loss"] = sft_loss
+            info_dict['loss_dict']["dpo_loss"] = preference_loss
+            info_dict['loss_dict']["dpo_acc"] = dpo_acc
+            info_dict['loss_dict']["chosen_reward"] = chosen_reward.mean()
+            info_dict['loss_dict']["reject_reward"] = reject_reward.mean()
+    return info_dict
+
+
+def batch_backward(model, scaler, info_dict):
+    if info_dict["train_engine"] == "deepspeed":
+        scaled_loss = model.backward(info_dict['loss_dict']['loss'])
+    else:
+        scaled_loss = info_dict['loss_dict']['loss'] / info_dict['accum_grad']
+        if scaler is not None:
+            scaler.scale(scaled_loss).backward()
+        else:
+            scaled_loss.backward()
+
+    info_dict['loss_dict']['loss'] = scaled_loss
+    return info_dict
+
+
+def update_parameter_and_lr(model, optimizer, scheduler, scaler, info_dict):
+    grad_norm = 0.0
+    if info_dict['train_engine'] == "deepspeed":
+        info_dict["is_gradient_accumulation_boundary"] = model.is_gradient_accumulation_boundary()
+        model.step()
+        grad_norm = model.get_global_grad_norm()
+    elif (info_dict['batch_idx'] + 1) % info_dict["accum_grad"] == 0:
+        # Use mixed precision training
+        if scaler is not None:
+            scaler.unscale_(optimizer)
+            grad_norm = clip_grad_norm_(model.parameters(), info_dict['grad_clip'])
+            # We don't check grad here since that if the gradient
+            # has inf/nan values, scaler.step will skip
+            # optimizer.step().
+            if torch.isfinite(grad_norm):
+                scaler.step(optimizer)
+            scaler.update()
+        else:
+            grad_norm = clip_grad_norm_(model.parameters(), info_dict['grad_clip'])
+            if torch.isfinite(grad_norm):
+                optimizer.step()
+        optimizer.zero_grad()
+        scheduler.step()
+    info_dict["lr"] = optimizer.param_groups[0]['lr']
+    info_dict["grad_norm"] = grad_norm
+    return info_dict
+
+
+def log_per_step(writer, info_dict):
+    tag = info_dict["tag"]
+    epoch = info_dict.get('epoch', 0)
+    step = info_dict["step"]
+    batch_idx = info_dict["batch_idx"]
+    loss_dict = info_dict['loss_dict']
+    rank = int(os.environ.get('RANK', 0))
+
+    # only rank 0 write to tensorboard to avoid multi-process write
+    if writer is not None:
+        if (info_dict['train_engine'] == 'deepspeed' and info_dict['is_gradient_accumulation_boundary'] is True) or \
+           (info_dict['train_engine'] == 'torch_ddp' and (info_dict['batch_idx'] + 1) % info_dict['accum_grad'] == 0):
+            for k in ['epoch', 'lr', 'grad_norm']:
+                writer.add_scalar('{}/{}'.format(tag, k), info_dict[k], step + 1)
+            for k, v in loss_dict.items():
+                writer.add_scalar('{}/{}'.format(tag, k), v, step + 1)
+
+    # TRAIN & CV, Shell log (stdout)
+    if (info_dict['batch_idx'] + 1) % info_dict['log_interval'] == 0:
+        log_str = '{} Batch {}/{} '.format(tag, epoch, batch_idx + 1)
+        for name, value in loss_dict.items():
+            log_str += '{} {:.6f} '.format(name, value)
+        if tag == "TRAIN":
+            log_str += 'lr {:.8f} grad_norm {:.6f}'.format(
+                info_dict["lr"], info_dict['grad_norm'])
+        log_str += ' rank {}'.format(rank)
+        logging.debug(log_str)
+
+
+def log_per_save(writer, info_dict):
+    tag = info_dict["tag"]
+    epoch = info_dict["epoch"]
+    step = info_dict["step"]
+    loss_dict = info_dict["loss_dict"]
+    lr = info_dict['lr']
+    rank = int(os.environ.get('RANK', 0))
+    logging.info(
+        'Epoch {} Step {} CV info lr {} {} rank {}'.format(
+            epoch, step + 1, lr, rank, ' '.join(['{}_{}'.format(k, v) for k, v in loss_dict.items()])))
+
+    if writer is not None:
+        for k in ['epoch', 'lr']:
+            writer.add_scalar('{}/{}'.format(tag, k), info_dict[k], step + 1)
+        for k, v in loss_dict.items():
+            writer.add_scalar('{}/{}'.format(tag, k), v, step + 1)

docker/Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04
+FROM nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04
 
 ARG VENV_NAME="cosyvoice"
 ENV VENV=$VENV_NAME

examples/libritts/cosyvoice/conf/cosyvoice_dpo.yaml

@@ -0,0 +1,226 @@
+# set random seed, so that you may reproduce your result.
+__set_seed1: !apply:random.seed [1986]
+__set_seed2: !apply:numpy.random.seed [1986]
+__set_seed3: !apply:torch.manual_seed [1986]
+__set_seed4: !apply:torch.cuda.manual_seed_all [1986]
+
+# fixed params
+sample_rate: 24000   # 16000 for llm, 24000 for cfm
+llm_input_size: 896
+llm_output_size: 896
+spk_embed_dim: 192
+qwen_pretrain_path: 'CosyVoice2-0.5B/CosyVoice-BlankEN'
+
+# model params
+# for all class/function included in this repo, we use !<name> or !<new> for intialization, so that user may find all corresponding class/function according to one single yaml.
+# for system/third_party class/function, we do not require this.
+llm: !new:cosyvoice.llm.llm_dpo.Qwen2LM
+    llm_input_size: !ref <llm_input_size>
+    llm_output_size: !ref <llm_output_size>
+    speech_token_size: 6561
+    length_normalized_loss: True
+    lsm_weight: 0
+    dpo: True
+    llm: !new:cosyvoice.llm.llm.Qwen2Encoder
+        pretrain_path: !ref <qwen_pretrain_path>
+    sampling: !name:cosyvoice.utils.common.ras_sampling
+        top_p: 0.8
+        top_k: 25
+        win_size: 10
+        tau_r: 0.1
+flow: !new:cosyvoice.flow.flow.CausalMaskedDiffWithXvec
+    input_size: 512
+    output_size: 80
+    spk_embed_dim: !ref <spk_embed_dim>
+    output_type: 'mel'
+    vocab_size: 6561
+    input_frame_rate: 25
+    only_mask_loss: True
+    token_mel_ratio: 2
+    pre_lookahead_len: 3
+    encoder: !new:cosyvoice.transformer.upsample_encoder.UpsampleConformerEncoder
+        output_size: 512
+        attention_heads: 8
+        linear_units: 2048
+        num_blocks: 6
+        dropout_rate: 0.1
+        positional_dropout_rate: 0.1
+        attention_dropout_rate: 0.1
+        normalize_before: True
+        input_layer: 'linear'
+        pos_enc_layer_type: 'rel_pos_espnet'
+        selfattention_layer_type: 'rel_selfattn'
+        input_size: 512
+        use_cnn_module: False
+        macaron_style: False
+    decoder: !new:cosyvoice.flow.flow_matching.CausalConditionalCFM
+        in_channels: 240
+        n_spks: 1
+        spk_emb_dim: 80
+        cfm_params: !new:omegaconf.DictConfig
+            content:
+                sigma_min: 1e-06
+                solver: 'euler'
+                t_scheduler: 'cosine'
+                training_cfg_rate: 0.2
+                inference_cfg_rate: 0.7
+                reg_loss_type: 'l1'
+        estimator: !new:cosyvoice.flow.decoder.ConditionalDecoder
+            in_channels: 320
+            out_channels: 80
+            causal: True
+            channels: [256]
+            dropout: 0.0
+            attention_head_dim: 64
+            n_blocks: 4
+            num_mid_blocks: 12
+            num_heads: 8
+            act_fn: 'gelu'
+
+hift: !new:cosyvoice.hifigan.generator.HiFTGenerator
+    in_channels: 80
+    base_channels: 512
+    nb_harmonics: 8
+    sampling_rate: !ref <sample_rate>
+    nsf_alpha: 0.1
+    nsf_sigma: 0.003
+    nsf_voiced_threshold: 10
+    upsample_rates: [8, 5, 3]
+    upsample_kernel_sizes: [16, 11, 7]
+    istft_params:
+        n_fft: 16
+        hop_len: 4
+    resblock_kernel_sizes: [3, 7, 11]
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]]
+    source_resblock_kernel_sizes: [7, 7, 11]
+    source_resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]]
+    lrelu_slope: 0.1
+    audio_limit: 0.99
+    f0_predictor: !new:cosyvoice.hifigan.f0_predictor.ConvRNNF0Predictor
+        num_class: 1
+        in_channels: 80
+        cond_channels: 512
+
+# gan related module
+mel_spec_transform1: !name:matcha.utils.audio.mel_spectrogram
+    n_fft: 1024
+    num_mels: 80
+    sampling_rate: !ref <sample_rate>
+    hop_size: 256
+    win_size: 1024
+    fmin: 0
+    fmax: null
+    center: False
+hifigan: !new:cosyvoice.hifigan.hifigan.HiFiGan
+    generator: !ref <hift>
+    discriminator: !new:cosyvoice.hifigan.discriminator.MultipleDiscriminator
+        mpd: !new:matcha.hifigan.models.MultiPeriodDiscriminator
+        mrd: !new:cosyvoice.hifigan.discriminator.MultiResolutionDiscriminator
+    mel_spec_transform: [
+        !ref <mel_spec_transform1>
+    ]
+
+# processor functions
+parquet_opener: !name:cosyvoice.dataset.processor.parquet_opener
+get_tokenizer: !name:whisper.tokenizer.get_tokenizer # change to !name:cosyvoice.tokenizer.tokenizer.get_tokenizer if you want to train with CosyVoice-300M-25Hz recipe
+    multilingual: True
+    num_languages: 100
+    language: 'en'
+    task: 'transcribe'
+allowed_special: 'all'
+tokenize: !name:cosyvoice.dataset.processor.tokenize
+    get_tokenizer: !ref <get_tokenizer>
+    allowed_special: !ref <allowed_special>
+filter: !name:cosyvoice.dataset.processor.filter
+    max_length: 40960
+    min_length: 0
+    token_max_length: 200
+    token_min_length: 1
+resample: !name:cosyvoice.dataset.processor.resample
+    resample_rate: !ref <sample_rate>
+truncate: !name:cosyvoice.dataset.processor.truncate
+    truncate_length: 24576 # must be a multiplier of hop_size
+feat_extractor: !name:matcha.utils.audio.mel_spectrogram
+    n_fft: 1024
+    num_mels: 80
+    sampling_rate: !ref <sample_rate>
+    hop_size: 256
+    win_size: 1024
+    fmin: 0
+    fmax: 8000
+    center: False
+compute_fbank: !name:cosyvoice.dataset.processor.compute_fbank
+    feat_extractor: !ref <feat_extractor>
+compute_f0: !name:cosyvoice.dataset.processor.compute_f0
+    sample_rate: !ref <sample_rate>
+    hop_size: 256
+parse_embedding: !name:cosyvoice.dataset.processor.parse_embedding
+    normalize: True
+shuffle: !name:cosyvoice.dataset.processor.shuffle
+    shuffle_size: 1000
+sort: !name:cosyvoice.dataset.processor.sort
+    sort_size: 500  # sort_size should be less than shuffle_size
+batch: !name:cosyvoice.dataset.processor.batch
+    batch_type: 'dynamic'
+    max_frames_in_batch: 2000 # change to 1400 in gan train on v100 16g
+padding: !name:cosyvoice.dataset.processor.padding
+    use_spk_embedding: True # change to True during sft
+    dpo: True
+
+# dataset processor pipeline
+data_pipeline: [
+    !ref <parquet_opener>,
+    !ref <tokenize>,
+    !ref <filter>,
+    !ref <resample>,
+    !ref <compute_fbank>,
+    !ref <parse_embedding>,
+    !ref <shuffle>,
+    !ref <sort>,
+    !ref <batch>,
+    !ref <padding>,
+]
+data_pipeline_gan: [
+    !ref <parquet_opener>,
+    !ref <tokenize>,
+    !ref <filter>,
+    !ref <resample>,
+    !ref <truncate>,
+    !ref <compute_fbank>,
+    !ref <compute_f0>,
+    !ref <parse_embedding>,
+    !ref <shuffle>,
+    !ref <sort>,
+    !ref <batch>,
+    !ref <padding>,
+]
+
+# llm flow train conf
+train_conf:
+    optim: adam
+    optim_conf:
+        lr: 0.00001 # change to 1e-5 during sft
+    scheduler: warmuplr # change to constantlr during sft
+    scheduler_conf:
+        warmup_steps: 25000
+    max_epoch: 200
+    grad_clip: 5
+    accum_grad: 2
+    log_interval: 100
+    save_per_step: -1
+
+# gan train conf
+train_conf_gan:
+    optim: adam
+    optim_conf:
+        lr: 0.0002 # use small lr for gan training
+    scheduler: constantlr
+    optim_d: adam
+    optim_conf_d:
+        lr: 0.0002 # use small lr for gan training
+    scheduler_d: constantlr
+    max_epoch: 200
+    grad_clip: 5
+    accum_grad: 1 # in gan training, accum_grad must be 1
+    log_interval: 100
+    save_per_step: -1

requirements.txt

@@ -18,7 +18,7 @@ networkx==3.1
 omegaconf==2.3.0
 onnx==1.16.0
 onnxruntime-gpu==1.18.0; sys_platform == 'linux'
-onnxruntime==1.18.0; sys_platform == 'darwin' or sys_platform == 'windows'
+onnxruntime==1.18.0; sys_platform == 'darwin' or sys_platform == 'win32'
 openai-whisper==20231117
 protobuf==4.25
 pyarrow==18.1.0

runtime/python/fastapi/server.py

@@ -72,6 +72,14 @@ async def inference_instruct(tts_text: str = Form(), spk_id: str = Form(), instr
     model_output = cosyvoice.inference_instruct(tts_text, spk_id, instruct_text)
     return StreamingResponse(generate_data(model_output))
 
+@app.get("/inference_instruct2")
+@app.post("/inference_instruct2")
+async def inference_instruct2(tts_text: str = Form(), instruct_text: str = Form(), prompt_wav: UploadFile = File()):
+    prompt_speech_16k = load_wav(prompt_wav.file, 16000)
+    model_output = cosyvoice.inference_instruct2(tts_text, instruct_text, prompt_speech_16k)
+    return StreamingResponse(generate_data(model_output))
+
+
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
@@ -90,4 +98,4 @@ if __name__ == '__main__':
             cosyvoice = CosyVoice2(args.model_dir)
         except Exception:
             raise TypeError('no valid model_type!')
-    uvicorn.run(app, host="0.0.0.0", port=args.port)
+    uvicorn.run(app, host="0.0.0.0", port=args.port)

tools/extract_speech_token.py

@@ -27,6 +27,9 @@ def single_job(utt):
     audio, sample_rate = torchaudio.load(utt2wav[utt], backend='soundfile')
     if sample_rate != 16000:
         audio = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(audio)
+    # Convert audio to mono
+    if audio.shape[0] > 1:
+        audio = audio.mean(dim=0, keepdim=True)
     if audio.shape[1] / 16000 > 30:
         logging.warning('do not support extract speech token for audio longer than 30s')
         speech_token = []

tools/make_parquet_list_dpo.py

@@ -0,0 +1,125 @@
+#!/usr/bin/env python3
+# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu)
+#
+# 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 argparse
+import logging
+import os
+import json
+from tqdm import tqdm
+import pandas as pd
+import multiprocessing
+import time
+import torch
+
+
+def job(utt_list, parquet_file, utt2parquet_file, spk2parquet_file):
+    start_time = time.time()
+    data_list = []
+    for utt in tqdm(utt_list):
+        data = open(utt2wav[utt], 'rb').read()
+        data_list.append(data)
+    wav_list = [utt2wav[utt] for utt in utt_list]
+    text_list = [utt2text[utt] for utt in utt_list]
+    spk_list = [utt2spk[utt] for utt in utt_list]
+    uttembedding_list = [utt2embedding[utt] for utt in utt_list]
+    spkembedding_list = [spk2embedding[utt2spk[utt]] for utt in utt_list]
+    speech_token_list = [utt2speech_token[utt] for utt in utt_list]
+    if utt2reject_speech_token:
+        reject_speech_token_list = [utt2reject_speech_token[utt] for utt in utt_list]
+
+    # 保存到parquet,utt2parquet_file,spk2parquet_file
+    df = pd.DataFrame()
+    df['utt'] = utt_list
+    df['wav'] = wav_list
+    df['audio_data'] = data_list
+    df['text'] = text_list
+    df['spk'] = spk_list
+    df['utt_embedding'] = uttembedding_list
+    df['spk_embedding'] = spkembedding_list
+    df['speech_token'] = speech_token_list
+    if utt2reject_speech_token:
+        df['reject_speech_token'] = reject_speech_token_list
+    df.to_parquet(parquet_file)
+    with open(utt2parquet_file, 'w') as f:
+        json.dump({k: parquet_file for k in utt_list}, f, ensure_ascii=False, indent=2)
+    with open(spk2parquet_file, 'w') as f:
+        json.dump({k: parquet_file for k in list(set(spk_list))}, f, ensure_ascii=False, indent=2)
+    logging.info('spend time {}'.format(time.time() - start_time))
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--num_utts_per_parquet',
+                        type=int,
+                        default=1000,
+                        help='num utts per parquet')
+    parser.add_argument('--num_processes',
+                        type=int,
+                        default=1,
+                        help='num processes for make parquets')
+    parser.add_argument('--src_dir',
+                        type=str)
+    parser.add_argument('--des_dir',
+                        type=str)
+    parser.add_argument('--dpo',
+                        action='store_true',
+                        default=False,
+                        help='Use Direct Preference Optimization')
+    args = parser.parse_args()
+
+    utt2wav, utt2text, utt2spk = {}, {}, {}
+    with open('{}/wav.scp'.format(args.src_dir)) as f:
+        for l in f:
+            l = l.replace('\n', '').split()
+            utt2wav[l[0]] = l[1]
+    with open('{}/text'.format(args.src_dir)) as f:
+        for l in f:
+            l = l.replace('\n', '').split()
+            utt2text[l[0]] = ' '.join(l[1:])
+    with open('{}/utt2spk'.format(args.src_dir)) as f:
+        for l in f:
+            l = l.replace('\n', '').split()
+            utt2spk[l[0]] = l[1]
+    utt2embedding = torch.load('{}/utt2embedding.pt'.format(args.src_dir))
+    spk2embedding = torch.load('{}/spk2embedding.pt'.format(args.src_dir))
+    utt2speech_token = torch.load('{}/utt2speech_token.pt'.format(args.src_dir))
+    if args.dpo:
+        utt2reject_speech_token = torch.load('{}/utt2reject_speech_token.pt'.format(args.src_dir))
+    else:
+        utt2reject_speech_token = None
+    utts = list(utt2wav.keys())
+
+    # Using process pool to speedup
+    pool = multiprocessing.Pool(processes=args.num_processes)
+    parquet_list, utt2parquet_list, spk2parquet_list = [], [], []
+    for i, j in enumerate(range(0, len(utts), args.num_utts_per_parquet)):
+        parquet_file = os.path.join(args.des_dir, 'parquet_{:09d}.tar'.format(i))
+        utt2parquet_file = os.path.join(args.des_dir, 'utt2parquet_{:09d}.json'.format(i))
+        spk2parquet_file = os.path.join(args.des_dir, 'spk2parquet_{:09d}.json'.format(i))
+        parquet_list.append(parquet_file)
+        utt2parquet_list.append(utt2parquet_file)
+        spk2parquet_list.append(spk2parquet_file)
+        pool.apply_async(job, (utts[j: j + args.num_utts_per_parquet], parquet_file, utt2parquet_file, spk2parquet_file))
+    pool.close()
+    pool.join()
+
+    with open('{}/data.list'.format(args.des_dir), 'w', encoding='utf8') as f1, \
+            open('{}/utt2data.list'.format(args.des_dir), 'w', encoding='utf8') as f2, \
+            open('{}/spk2data.list'.format(args.des_dir), 'w', encoding='utf8') as f3:
+        for name in parquet_list:
+            f1.write(name + '\n')
+        for name in utt2parquet_list:
+            f2.write(name + '\n')
+        for name in spk2parquet_list:
+            f3.write(name + '\n')