init zipformer_llm_zh

egs/speech_llm/ASR_LLM/zipformer_llm_zh/asr_datamodule.py

@@ -0,0 +1 @@
+../whisper_llm_zh/asr_datamodule.py

egs/speech_llm/ASR_LLM/zipformer_llm_zh/decode.py

@@ -0,0 +1,608 @@
+#!/usr/bin/env python3
+# Copyright 2021 Xiaomi Corporation (Author: Liyong Guo,
+#                                            Fangjun Kuang,
+#                                            Wei Kang)
+#           2024 Yuekai Zhang
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# 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.
+"""
+Usage:
+# Command for decoding using fine-tuned models:
+
+pip install huggingface_hub['cli']
+mkdir -p models/whisper models/qwen models/checkpoint
+huggingface-cli download --local-dir models/checkpoint yuekai/icefall_asr_aishell_whisper_qwen2_1.5B
+
+# For aishell fine-tuned whisper model
+huggingface-cli download --local-dir models/whisper    yuekai/icefall_asr_aishell_whisper exp_large_v2/whisper-large-v2-aishell1-epoch-10-avg-6.pt
+# For multi-hans fine-tuned whisper model
+# huggingface-cli download --local-dir models/whisper    yuekai/icefall_asr_multi-hans-zh_whisper v1.1/whisper-large-v2-multi-hans-zh-epoch-3-avg-10.pt
+
+huggingface-clie download  --local-dir models/qwen     Qwen/Qwen2-7B-Instruct
+
+mkdir -p whisper_llm_zh/exp_aishell_whisper_qwen2_1.5B
+ln -s models/checkpoint/epoch-10-avg-5.pt whisper_llm_zh/exp_aishell_whisper_qwen2_1.5B/epoch-999.pt
+
+python3 ./whisper_llm_zh/decode.py \
+  --max-duration 80 \
+  --exp-dir whisper_llm_zh/exp_aishell_whisper_qwen2_1.5B \
+  --speech-encoder-path-or-name models/whisper/exp_large_v2/whisper-large-v2-aishell1-epoch-10-avg-6.pt  \
+  --llm-path-or-name models/qwen \
+  --epoch 999 --avg 1 \
+  --manifest-dir data/fbank \
+  --use-flash-attn True \
+  --use-lora True --dataset aishell
+"""
+
+import argparse
+import logging
+from collections import defaultdict
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple
+
+import k2
+import torch
+import torch.nn as nn
+import transformers
+import whisper
+from asr_datamodule import AsrDataModule
+from lhotse.cut import Cut
+from model import SPEECH_LLM, EncoderProjector
+from multi_dataset import MultiDataset
+from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training
+from train import DEFAULT_SPEECH_TOKEN
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from whisper_encoder_forward_monkey_patch import replace_whisper_encoder_forward
+
+from icefall.checkpoint import load_checkpoint
+from icefall.env import get_env_info
+from icefall.utils import (
+    AttributeDict,
+    setup_logger,
+    store_transcripts,
+    str2bool,
+    write_error_stats,
+)
+
+
+def average_checkpoints(
+    filenames: List[Path], device: torch.device = torch.device("cpu")
+) -> dict:
+    """Average a list of checkpoints.
+    The function is mainly used for deepspeed converted checkpoint averaging, which only include model state_dict.
+
+    Args:
+      filenames:
+        Filenames of the checkpoints to be averaged. We assume all
+        checkpoints are saved by :func:`save_checkpoint`.
+      device:
+        Move checkpoints to this device before averaging.
+    Returns:
+      Return a dict (i.e., state_dict) which is the average of all
+      model state dicts contained in the checkpoints.
+    """
+    n = len(filenames)
+
+    if "model" in torch.load(filenames[0], map_location=device):
+        avg = torch.load(filenames[0], map_location=device)["model"]
+    else:
+        avg = torch.load(filenames[0], map_location=device)
+
+    # Identify shared parameters. Two parameters are said to be shared
+    # if they have the same data_ptr
+    uniqued: Dict[int, str] = dict()
+
+    for k, v in avg.items():
+        v_data_ptr = v.data_ptr()
+        if v_data_ptr in uniqued:
+            continue
+        uniqued[v_data_ptr] = k
+
+    uniqued_names = list(uniqued.values())
+
+    for i in range(1, n):
+        if "model" in torch.load(filenames[i], map_location=device):
+            state_dict = torch.load(filenames[i], map_location=device)["model"]
+        else:
+            state_dict = torch.load(filenames[i], map_location=device)
+        for k in uniqued_names:
+            avg[k] += state_dict[k]
+
+    for k in uniqued_names:
+        if avg[k].is_floating_point():
+            avg[k] /= n
+        else:
+            avg[k] //= n
+
+    return avg
+
+
+def add_model_arguments(parser: argparse.ArgumentParser):
+    parser.add_argument(
+        "--llm-path-or-name",
+        type=str,
+        default="/workspace/asr/Qwen1.5-0.5B-Chat",
+        help="Path or name of the large language model.",
+    )
+
+    parser.add_argument(
+        "--speech-encoder-path-or-name",
+        type=str,
+        default="whisper-large-v2",
+        help="Path or name of the speech encoder.",
+    )
+
+    parser.add_argument(
+        "--encoder-projector-ds-rate",
+        type=int,
+        default=8,
+        help="Downsample rate for the encoder projector.",
+    )
+
+    parser.add_argument(
+        "--use-flash-attn",
+        type=str2bool,
+        default=True,
+        help="Whether to use flash attention.",
+    )
+
+    parser.add_argument(
+        "--use-lora",
+        type=str2bool,
+        default=True,
+        help="Whether to use lora fine-tuned llm checkpoint.",
+    )
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--epoch",
+        type=int,
+        default=-1,
+        help="It specifies the checkpoint to use for decoding."
+        "Note: Epoch counts from 0.",
+    )
+    parser.add_argument(
+        "--avg",
+        type=int,
+        default=1,
+        help="Number of checkpoints to average. Automatically select "
+        "consecutive checkpoints before the checkpoint specified by "
+        "'--epoch'. ",
+    )
+
+    parser.add_argument(
+        "--method",
+        type=str,
+        default="beam-search",
+        help="""Decoding method.
+        Supported values are:
+          - beam-search
+        """,
+    )
+
+    parser.add_argument(
+        "--beam-size",
+        type=int,
+        default=1,
+        help="beam size for beam search decoding",
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="whisper/exp",
+        help="The experiment dir",
+    )
+
+    parser.add_argument(
+        "--remove-whisper-encoder-input-length-restriction",
+        type=str2bool,
+        default=True,
+        help="replace whisper encoder forward method to remove input length restriction",
+    )
+
+    parser.add_argument(
+        "--dataset",
+        type=str,
+        default="aishell",
+        choices=["aishell", "speechio", "wenetspeech_test_meeting", "multi_hans_zh"],
+        help="The dataset to decode",
+    )
+
+    add_model_arguments(parser)
+    return parser
+
+
+def get_params() -> AttributeDict:
+    params = AttributeDict(
+        {
+            "env_info": get_env_info(),
+        }
+    )
+    return params
+
+
+def decode_one_batch(
+    params: AttributeDict,
+    model: nn.Module,
+    tokenizer: AutoTokenizer,
+    batch: dict,
+) -> Dict[str, List[List[int]]]:
+    """Decode one batch and return the result in a dict. The dict has the
+    following format:
+
+        - key: "beam-search"
+        - value: A list of lists. Each sublist is a list of token IDs.
+    Args:
+        params:
+            It is returned by :func:`get_params`.
+        model:
+            The neural model.
+        batch:
+            It is returned by :meth:`torch.utils.data.DataLoader.__iter__`.
+    Returns:
+        Return a dict, whose key may be "beam-search".
+    """
+
+    def preprocess(
+        messages,
+        tokenizer: transformers.PreTrainedTokenizer,
+        max_len: int = 128,
+    ) -> Dict:
+        """Preprocesses the data for supervised fine-tuning."""
+        texts = []
+        TEMPLATE = "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content']}}{% if loop.last %}{{''}}{% else %}{{ '<|im_end|>\n' }}{% endif %}{% endfor %}"
+        for i, msg in enumerate(messages):
+            texts.append(
+                tokenizer.apply_chat_template(
+                    msg,
+                    tokenize=True,
+                    add_generation_prompt=False,
+                    chat_template=TEMPLATE,
+                    padding="longest",
+                    max_length=max_len,
+                    truncation=True,
+                )
+            )
+        max_len_texts = max([len(text) for text in texts])
+        if tokenizer.padding_side == "right":
+            texts = [
+                text + [tokenizer.pad_token_id] * (max_len_texts - len(text))
+                for text in texts
+            ]
+        else:
+            texts = [
+                [tokenizer.pad_token_id] * (max_len_texts - len(text)) + text
+                for text in texts
+            ]
+
+        input_ids = torch.tensor(texts, dtype=torch.int)
+
+        attention_mask = input_ids.ne(tokenizer.pad_token_id)
+
+        return input_ids, attention_mask
+
+    dtype = torch.float32
+    device = model.llm.device
+
+    feature = batch["inputs"]
+    assert feature.ndim == 3
+    feature = feature.to(device, dtype=dtype).transpose(1, 2)
+    if not params.remove_whisper_encoder_input_length_restriction:
+        T = 3000
+        if feature.shape[2] < T:
+            feature = torch.cat(
+                [
+                    feature,
+                    torch.zeros(
+                        feature.shape[0], feature.shape[1], T - feature.shape[2]
+                    ).to(device, dtype=dtype),
+                ],
+                2,
+            )
+
+    supervisions = batch["supervisions"]
+    feature_len = supervisions["num_frames"]
+    feature_len = feature_len.to(device, dtype=dtype)
+
+    messages = [
+        [
+            {"role": "user", "content": f"{DEFAULT_SPEECH_TOKEN}请转写音频为文字"},
+            {"role": "assistant", "content": ""},
+        ]
+    ] * len(feature)
+
+    input_ids, attention_mask = preprocess(messages, tokenizer, max_len=128)
+
+    generated_ids = model.decode(
+        feature, input_ids.to(device, dtype=torch.long), attention_mask.to(device)
+    )
+    hyps = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
+
+    return {"beam-search": hyps}
+
+
+def decode_dataset(
+    dl: torch.utils.data.DataLoader,
+    params: AttributeDict,
+    model: nn.Module,
+    tokenizer: AutoTokenizer,
+) -> Dict[str, List[Tuple[str, List[str], List[str]]]]:
+    """Decode dataset.
+
+    Args:
+        dl:
+            The dataloader.
+        params:
+            It is returned by :func:`get_params`.
+        model:
+            The neural model.
+    Returns:
+        Return a dict, whose key may be "beam-search".
+    """
+    results = []
+
+    num_cuts = 0
+
+    try:
+        num_batches = len(dl)
+    except TypeError:
+        num_batches = "?"
+
+    results = defaultdict(list)
+    for batch_idx, batch in enumerate(dl):
+        texts = batch["supervisions"]["text"]
+        texts = [list("".join(text.split())) for text in texts]
+        cut_ids = [cut.id for cut in batch["supervisions"]["cut"]]
+
+        hyps_dict = decode_one_batch(
+            params=params,
+            model=model,
+            batch=batch,
+            tokenizer=tokenizer,
+        )
+
+        for lm_scale, hyps in hyps_dict.items():
+            this_batch = []
+            assert len(hyps) == len(texts)
+            for cut_id, hyp_text, ref_text in zip(cut_ids, hyps, texts):
+                this_batch.append((cut_id, ref_text, hyp_text))
+
+            results[lm_scale].extend(this_batch)
+
+        num_cuts += len(batch["supervisions"]["text"])
+
+        if batch_idx % 100 == 0:
+            batch_str = f"{batch_idx}/{num_batches}"
+
+            logging.info(f"batch {batch_str}, cuts processed until now is {num_cuts}")
+    return results
+
+
+def save_results(
+    params: AttributeDict,
+    test_set_name: str,
+    results_dict: Dict[str, List[Tuple[List[int], List[int]]]],
+):
+    test_set_wers = dict()
+    for key, results in results_dict.items():
+        recog_path = (
+            params.res_dir / f"recogs-{test_set_name}-{key}-{params.suffix}.txt"
+        )
+        results = sorted(results)
+        store_transcripts(filename=recog_path, texts=results, char_level=True)
+        logging.info(f"The transcripts are stored in {recog_path}")
+
+        # The following prints out CERs, per-word error statistics and aligned
+        # ref/hyp pairs.
+        errs_filename = (
+            params.res_dir / f"errs-{test_set_name}-{key}-{params.suffix}.txt"
+        )
+        with open(errs_filename, "w") as f:
+            wer = write_error_stats(
+                f,
+                f"{test_set_name}-{key}",
+                results,
+                enable_log=True,
+                compute_CER=True,
+            )
+            test_set_wers[key] = wer
+
+        logging.info("Wrote detailed error stats to {}".format(errs_filename))
+
+    test_set_wers = sorted(test_set_wers.items(), key=lambda x: x[1])
+    errs_info = (
+        params.res_dir / f"wer-summary-{test_set_name}-{key}-{params.suffix}.txt"
+    )
+    with open(errs_info, "w") as f:
+        print("settings\tCER", file=f)
+        for key, val in test_set_wers:
+            print("{}\t{}".format(key, val), file=f)
+
+    s = "\nFor {}, CER of different settings are:\n".format(test_set_name)
+    note = "\tbest for {}".format(test_set_name)
+    for key, val in test_set_wers:
+        s += "{}\t{}{}\n".format(key, val, note)
+        note = ""
+    logging.info(s)
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    AsrDataModule.add_arguments(parser)
+    args = parser.parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    params.res_dir = params.exp_dir / f"{params.method}"
+
+    params.suffix = f"epoch-{params.epoch}-avg-{params.avg}"
+    setup_logger(
+        params.res_dir
+        / f"log-decode-{params.method}-beam{params.beam_size}-{params.suffix}"
+    )
+
+    logging.info("Decoding started")
+    logging.info(params)
+
+    device = torch.device("cpu")
+    if torch.cuda.is_available():
+        device = torch.device("cuda")
+
+    logging.info(f"device: {device}")
+
+    if params.remove_whisper_encoder_input_length_restriction:
+        replace_whisper_encoder_forward()
+
+    whisper_model = whisper.load_model(params.speech_encoder_path_or_name, "cpu")
+    speech_encoder = whisper_model.encoder
+    speech_encoder_dim = whisper_model.dims.n_audio_state
+    tokenizer = AutoTokenizer.from_pretrained(params.llm_path_or_name)
+
+    if params.use_flash_attn:
+        attn_implementation = "flash_attention_2"
+        # torch_dtype=torch.bfloat16 FIX ME
+        torch_dtype = torch.float16
+        tokenizer.padding_side = "left"
+
+    else:
+        attn_implementation = "eager"
+        torch_dtype = torch.float16
+        tokenizer.padding_side = "right"
+
+    llm = AutoModelForCausalLM.from_pretrained(
+        params.llm_path_or_name,
+        attn_implementation=attn_implementation,
+        torch_dtype=torch_dtype,
+    )
+    if params.use_lora:
+        lora_config = LoraConfig(
+            r=64,
+            lora_alpha=16,
+            target_modules=[
+                "q_proj",
+                "k_proj",
+                "v_proj",
+                "o_proj",
+                "up_proj",
+                "gate_proj",
+                "down_proj",
+            ],
+            task_type="CAUSAL_LM",
+        )
+        llm = get_peft_model(llm, lora_config)
+        llm.print_trainable_parameters()
+
+    special_tokens_dict = {"additional_special_tokens": [DEFAULT_SPEECH_TOKEN]}
+    tokenizer.add_special_tokens(special_tokens_dict)
+    llm.config.pad_token_id = tokenizer.convert_tokens_to_ids("<|endoftext|>")
+    llm.config.bos_token_id = tokenizer.convert_tokens_to_ids("<|im_start|>")
+    llm.config.eos_token_id = tokenizer.convert_tokens_to_ids("<|im_end|>")
+
+    llm.config.default_speech_token_id = tokenizer.convert_tokens_to_ids(
+        DEFAULT_SPEECH_TOKEN
+    )
+
+    encoder_projector = EncoderProjector(
+        speech_encoder_dim, llm.config.hidden_size, params.encoder_projector_ds_rate
+    )
+
+    model = SPEECH_LLM(
+        speech_encoder,
+        llm,
+        encoder_projector,
+    )
+
+    if params.avg > 1:
+        start = params.epoch - params.avg + 1
+        assert start >= 1, start
+        # deepspeed converted checkpoint only contains model state_dict
+        filenames = [
+            f"{params.exp_dir}/epoch-{epoch}/pytorch_model.bin"
+            for epoch in range(start, params.epoch + 1)
+        ]
+        avg_checkpoint = average_checkpoints(filenames)
+        model.load_state_dict(avg_checkpoint, strict=False)
+
+        # filename = f"{params.exp_dir}/epoch-{params.epoch}-avg-{params.avg}.pt"
+        # torch.save(avg_checkpoint, filename)
+    else:
+        checkpoint = torch.load(
+            f"{params.exp_dir}/epoch-{params.epoch}/pytorch_model.bin",
+            map_location="cpu",
+        )
+        model.load_state_dict(checkpoint, strict=False)
+
+    model.to(device)
+    model.eval()
+    num_param = sum([p.numel() for p in model.parameters()])
+    logging.info(f"Number of model parameters: {num_param}")
+
+    # we need cut ids to display recognition results.
+    args.return_cuts = True
+
+    data_module = AsrDataModule(args)
+    multi_dataset = MultiDataset(args.manifest_dir)
+
+    def remove_long_utt(c: Cut):
+        # Keep only utterances with duration in 30 seconds
+        #
+        if c.duration > 30.0:
+            logging.warning(
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
+            )
+            return False
+        return True
+
+    if params.dataset == "aishell":
+        test_sets_cuts = multi_dataset.aishell_test_cuts()
+    elif params.dataset == "speechio":
+        test_sets_cuts = multi_dataset.speechio_test_cuts()
+    elif params.dataset == "wenetspeech_test_meeting":
+        test_sets_cuts = multi_dataset.wenetspeech_test_meeting_cuts()
+    else:
+        test_sets_cuts = multi_dataset.test_cuts()
+
+    test_sets = test_sets_cuts.keys()
+    test_dls = [
+        data_module.test_dataloaders(test_sets_cuts[cuts_name].filter(remove_long_utt))
+        for cuts_name in test_sets
+    ]
+
+    for test_set, test_dl in zip(test_sets, test_dls):
+        results_dict = decode_dataset(
+            dl=test_dl,
+            params=params,
+            model=model,
+            tokenizer=tokenizer,
+        )
+
+        save_results(params=params, test_set_name=test_set, results_dict=results_dict)
+
+    logging.info("Done!")
+
+
+if __name__ == "__main__":
+    torch.set_num_threads(1)
+    torch.set_num_interop_threads(1)
+    main()

egs/speech_llm/ASR_LLM/zipformer_llm_zh/ds_config_zero1.json

@@ -0,0 +1 @@
+../whisper_llm_zh/ds_config_zero1.json

egs/speech_llm/ASR_LLM/zipformer_llm_zh/model.py

@@ -0,0 +1,285 @@
+import torch
+from torch import nn
+from transformers.trainer_pt_utils import LabelSmoother
+
+IGNORE_TOKEN_ID = LabelSmoother.ignore_index
+
+
+class EncoderProjector(nn.Module):
+    """
+    The encoder projector module. It is used to project the encoder outputs to the same dimension as the language model.
+    Modified from https://github.com/X-LANCE/SLAM-LLM/blob/main/src/slam_llm/models/projector.py.
+    Args:
+        encoder_dim (:obj:`int`): The dimension of the encoder outputs.
+        llm_dim (:obj:`int`): The dimension of the language model.
+        downsample_rate (:obj:`int`, `optional`, defaults to 5): The downsample rate to use.
+    """
+
+    def __init__(self, encoder_dim, llm_dim, downsample_rate=5):
+        super().__init__()
+        self.downsample_rate = downsample_rate
+        self.linear1 = nn.Linear(encoder_dim * self.downsample_rate, llm_dim)
+        self.relu = nn.ReLU()
+        self.linear2 = nn.Linear(llm_dim, llm_dim)
+
+    def forward(self, x):
+
+        batch_size, seq_len, feat_dim = x.size()
+        num_frames_to_discard = seq_len % self.downsample_rate
+        if num_frames_to_discard > 0:
+            x = x[:, :-num_frames_to_discard, :]
+        seq_len = x.size(1)
+
+        x = x.contiguous()
+        x = x.view(
+            batch_size, seq_len // self.downsample_rate, feat_dim * self.downsample_rate
+        )
+
+        x = self.linear1(x)
+        x = self.relu(x)
+        x = self.linear2(x)
+        return x
+
+
+class SPEECH_LLM(nn.Module):
+    """
+    The Speech-to-Text model. It consists of an encoder, a language model and an encoder projector.
+    The encoder is used to extract speech features from the input speech signal.
+    The encoder projector is used to project the encoder outputs to the same dimension as the language model.
+    The language model is used to generate the text from the speech features.
+    Args:
+        encoder (:obj:`nn.Module`): The encoder module.
+        llm (:obj:`nn.Module`): The language model module.
+        encoder_projector (:obj:`nn.Module`): The encoder projector module.
+    """
+
+    def __init__(
+        self,
+        encoder: nn.Module,
+        llm: nn.Module,
+        encoder_projector: nn.Module,
+    ):
+        super().__init__()
+        self.encoder = encoder
+        self.llm = llm
+        self.encoder_projector = encoder_projector
+
+    def _merge_input_ids_with_speech_features(
+        self, speech_features, inputs_embeds, input_ids, attention_mask, labels=None
+    ):
+        """
+        Merge the speech features with the input_ids and attention_mask. This is done by replacing the speech tokens
+        with the speech features and padding the input_ids to the maximum length of the speech features.
+        Modified from https://github.com/huggingface/transformers/blob/main/src/transformers/models/llava/modeling_llava.py#L277.
+        Args:
+            speech_features (:obj:`torch.Tensor`): The speech features to merge with the input_ids.
+            inputs_embeds (:obj:`torch.Tensor`): The embeddings of the input_ids.
+            input_ids (:obj:`torch.Tensor`): The input ids to merge.
+            attention_mask (:obj:`torch.Tensor`): The attention mask to merge.
+            labels (:obj:`torch.Tensor`, `optional`): The labels to merge.
+        Returns:
+            :obj:`Tuple(torch.Tensor)`: The merged embeddings, attention mask, labels and position ids.
+        """
+        num_speechs, speech_len, embed_dim = speech_features.shape
+        batch_size, sequence_length = input_ids.shape
+        left_padding = not torch.sum(
+            input_ids[:, -1] == torch.tensor(self.llm.config.pad_token_id)
+        )
+        # 1. Create a mask to know where special speech tokens are
+        special_speech_token_mask = input_ids == self.llm.config.default_speech_token_id
+        num_special_speech_tokens = torch.sum(special_speech_token_mask, dim=-1)
+        # Compute the maximum embed dimension
+        max_embed_dim = (
+            num_special_speech_tokens.max() * (speech_len - 1)
+        ) + sequence_length
+        batch_indices, non_speech_indices = torch.where(
+            input_ids != self.llm.config.default_speech_token_id
+        )
+
+        # 2. Compute the positions where text should be written
+        # Calculate new positions for text tokens in merged speech-text sequence.
+        # `special_speech_token_mask` identifies speech tokens. Each speech token will be replaced by `nb_text_tokens_per_speechs - 1` text tokens.
+        # `torch.cumsum` computes how each speech token shifts subsequent text token positions.
+        # - 1 to adjust for zero-based indexing, as `cumsum` inherently increases indices by one.
+        new_token_positions = (
+            torch.cumsum((special_speech_token_mask * (speech_len - 1) + 1), -1) - 1
+        )
+        nb_speech_pad = max_embed_dim - 1 - new_token_positions[:, -1]
+        if left_padding:
+            new_token_positions += nb_speech_pad[:, None]  # offset for left padding
+        text_to_overwrite = new_token_positions[batch_indices, non_speech_indices]
+
+        # 3. Create the full embedding, already padded to the maximum position
+        final_embedding = torch.zeros(
+            batch_size,
+            max_embed_dim,
+            embed_dim,
+            dtype=inputs_embeds.dtype,
+            device=inputs_embeds.device,
+        )
+        final_attention_mask = torch.zeros(
+            batch_size,
+            max_embed_dim,
+            dtype=attention_mask.dtype,
+            device=inputs_embeds.device,
+        )
+        if labels is not None:
+            final_labels = torch.full(
+                (batch_size, max_embed_dim),
+                IGNORE_TOKEN_ID,
+                dtype=input_ids.dtype,
+                device=input_ids.device,
+            )
+        # In case the Vision model or the Language model has been offloaded to CPU, we need to manually
+        # set the corresponding tensors into their correct target device.
+        target_device = inputs_embeds.device
+        batch_indices, non_speech_indices, text_to_overwrite = (
+            batch_indices.to(target_device),
+            non_speech_indices.to(target_device),
+            text_to_overwrite.to(target_device),
+        )
+        attention_mask = attention_mask.to(target_device)
+
+        # 4. Fill the embeddings based on the mask. If we have ["hey" "<speech>", "how", "are"]
+        # we need to index copy on [0, 577, 578, 579] for the text and [1:576] for the speech features
+        final_embedding[batch_indices, text_to_overwrite] = inputs_embeds[
+            batch_indices, non_speech_indices
+        ]
+        final_attention_mask[batch_indices, text_to_overwrite] = attention_mask[
+            batch_indices, non_speech_indices
+        ]
+        if labels is not None:
+            final_labels[batch_indices, text_to_overwrite] = labels[
+                batch_indices, non_speech_indices
+            ]
+
+        # 5. Fill the embeddings corresponding to the speechs. Anything that is not `text_positions` needs filling (#29835)
+        speech_to_overwrite = torch.full(
+            (batch_size, max_embed_dim),
+            True,
+            dtype=torch.bool,
+            device=inputs_embeds.device,
+        )
+        speech_to_overwrite[batch_indices, text_to_overwrite] = False
+        speech_to_overwrite &= speech_to_overwrite.cumsum(-1) - 1 >= nb_speech_pad[
+            :, None
+        ].to(target_device)
+
+        if speech_to_overwrite.sum() != speech_features.shape[:-1].numel():
+            raise ValueError(
+                f"The input provided to the model are wrong. The number of speech tokens is {torch.sum(special_speech_token_mask)} while"
+                f" the number of speech given to the model is {num_speechs}. This prevents correct indexing and breaks batch generation."
+            )
+
+        final_embedding[speech_to_overwrite] = (
+            speech_features.contiguous().reshape(-1, embed_dim).to(target_device)
+        )
+        final_attention_mask |= speech_to_overwrite
+        position_ids = (final_attention_mask.cumsum(-1) - 1).masked_fill_(
+            (final_attention_mask == 0), 1
+        )
+
+        # 6. Mask out the embedding at padding positions, as we later use the past_key_value value to determine the non-attended tokens.
+        batch_indices, pad_indices = torch.where(
+            input_ids == self.llm.config.pad_token_id
+        )
+        indices_to_mask = new_token_positions[batch_indices, pad_indices]
+
+        final_embedding[batch_indices, indices_to_mask] = 0
+
+        if labels is None:
+            final_labels = None
+
+        return final_embedding, final_attention_mask, final_labels, position_ids
+
+    def forward(
+        self,
+        fbank: torch.Tensor = None,
+        input_ids: torch.LongTensor = None,
+        attention_mask: torch.Tensor = None,
+        labels: torch.LongTensor = None,
+    ):
+        encoder_outs = self.encoder(fbank)
+
+        speech_features = self.encoder_projector(encoder_outs)
+
+        inputs_embeds = self.llm.get_input_embeddings()(input_ids)
+
+        (
+            inputs_embeds,
+            attention_mask,
+            labels,
+            _,
+        ) = self._merge_input_ids_with_speech_features(
+            speech_features, inputs_embeds, input_ids, attention_mask, labels
+        )
+
+        model_outputs = self.llm(
+            inputs_embeds=inputs_embeds, attention_mask=attention_mask, labels=labels
+        )
+
+        with torch.no_grad():
+            preds = torch.argmax(model_outputs.logits, -1)
+            acc = compute_accuracy(
+                preds.detach()[:, :-1],
+                labels.detach()[:, 1:],
+                ignore_label=IGNORE_TOKEN_ID,
+            )
+        return model_outputs, acc
+
+    def decode(
+        self,
+        fbank: torch.Tensor = None,
+        input_ids: torch.LongTensor = None,
+        attention_mask: torch.Tensor = None,
+        **kwargs,
+    ):
+
+        encoder_outs = self.encoder(fbank)
+        speech_features = self.encoder_projector(encoder_outs)
+        speech_features = speech_features.to(torch.float16)
+        inputs_embeds = self.llm.get_input_embeddings()(input_ids)
+        (
+            inputs_embeds,
+            attention_mask,
+            _,
+            position_ids,
+        ) = self._merge_input_ids_with_speech_features(
+            speech_features, inputs_embeds, input_ids, attention_mask
+        )
+        generated_ids = self.llm.generate(
+            inputs_embeds=inputs_embeds,
+            max_new_tokens=kwargs.get("max_new_tokens", 200),
+            num_beams=kwargs.get("num_beams", 1),
+            do_sample=kwargs.get("do_sample", False),
+            min_length=kwargs.get("min_length", 1),
+            top_p=kwargs.get("top_p", 1.0),
+            repetition_penalty=kwargs.get("repetition_penalty", 1.0),
+            length_penalty=kwargs.get("length_penalty", 1.0),
+            temperature=kwargs.get("temperature", 1.0),
+            bos_token_id=self.llm.config.bos_token_id,
+            eos_token_id=self.llm.config.eos_token_id,
+            pad_token_id=self.llm.config.pad_token_id,
+        )
+
+        return generated_ids
+
+
+def compute_accuracy(pad_outputs, pad_targets, ignore_label):
+    """Calculate accuracy.
+    Copied from https://github.com/X-LANCE/SLAM-LLM/blob/main/src/slam_llm/utils/metric.py
+    Args:
+        pad_outputs (LongTensor): Prediction tensors (B, Lmax).
+        pad_targets (LongTensor): Target label tensors (B, Lmax).
+        ignore_label (int): Ignore label id.
+
+    Returns:
+        float: Accuracy value (0.0 - 1.0).
+
+    """
+    mask = pad_targets != ignore_label
+    numerator = torch.sum(
+        pad_outputs.masked_select(mask) == pad_targets.masked_select(mask)
+    )
+    denominator = torch.sum(mask)
+    return numerator.float() / denominator.float()

egs/speech_llm/ASR_LLM/zipformer_llm_zh/multi_dataset.py

@@ -0,0 +1 @@
+../whisper_llm_zh/multi_dataset.py

egs/speech_llm/ASR_LLM/zipformer_llm_zh/train.py

@@ -0,0 +1,815 @@
+#!/usr/bin/env python3
+# Copyright    2023  Xiaomi Corp.        (authors: Xiaoyu Yang)
+#              2024  Yuekai Zhang
+#              2025  Yifan  Yang
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# 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.
+"""
+Usage:
+# fine-tuning with whisper and Qwen2
+pip install huggingface_hub['cli']
+mkdir -p models/whisper models/qwen
+
+# For aishell fine-tuned whisper model
+huggingface-cli download --local-dir models/whisper    yuekai/icefall_asr_aishell_whisper exp_large_v2/whisper-large-v2-aishell1-epoch-10-avg-6.pt
+# For multi-hans fine-tuned whisper model
+# huggingface-cli download --local-dir models/whisper    yuekai/icefall_asr_multi-hans-zh_whisper v1.1/whisper-large-v2-multi-hans-zh-epoch-3-avg-10.pt
+
+# huggingface-clie download  --local-dir models/qwen     Qwen/Qwen2-7B-Instruct
+huggingface-clie download  --local-dir models/qwen     Qwen/Qwen2-1.5B-Instruct
+
+torchrun --nproc_per_node 8 ./whisper_llm_zh/train.py \
+  --max-duration 200 \
+  --exp-dir ./whisper_llm_zh/exp_test \
+  --speech-encoder-path-or-name models/whisper/exp_large_v2/whisper-large-v2-aishell1-epoch-10-avg-6.pt \
+  --llm-path-or-name Qwen/Qwen2-1.5B-Instruct \
+  --manifest-dir data/fbank \
+  --deepspeed \
+  --deepspeed_config ./whisper_llm_zh/ds_config_zero1.json \
+  --use-flash-attn True \
+  --use-lora True --unfreeze-llm True
+"""
+
+import argparse
+import logging
+import os
+import warnings
+from pathlib import Path
+from typing import Dict, Optional, Tuple
+
+import deepspeed
+import torch
+import torch.nn as nn
+import transformers
+import whisper
+from asr_datamodule import AsrDataModule
+from deepspeed.utils.zero_to_fp32 import convert_zero_checkpoint_to_fp32_state_dict
+from lhotse.cut import Cut
+from lhotse.utils import fix_random_seed
+from model import IGNORE_TOKEN_ID, SPEECH_LLM, EncoderProjector
+from multi_dataset import MultiDataset
+from peft import LoraConfig, get_peft_model
+from torch import Tensor
+from torch.utils.tensorboard import SummaryWriter
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from whisper_encoder_forward_monkey_patch import replace_whisper_encoder_forward
+
+from icefall.dist import get_rank, get_world_size
+from icefall.env import get_env_info
+from icefall.utils import AttributeDict, MetricsTracker, setup_logger, str2bool
+
+DEFAULT_SPEECH_TOKEN = "<speech>"
+
+
+def set_batch_count(model: nn.Module, batch_count: float) -> None:
+    for module in model.modules():
+        if hasattr(module, "batch_count"):
+            module.batch_count = batch_count
+
+
+def add_model_arguments(parser: argparse.ArgumentParser):
+    parser.add_argument(
+        "--llm-path-or-name",
+        type=str,
+        default="/workspace/asr/Qwen1.5-0.5B-Chat",
+        help="Path or name of the large language model.",
+    )
+
+    parser.add_argument(
+        "--speech-encoder-path-or-name",
+        type=str,
+        default="whisper-large-v2",
+        help="Path or name of the speech encoder.",
+    )
+
+    parser.add_argument(
+        "--encoder-projector-ds-rate",
+        type=int,
+        default=8,
+        help="Downsample rate for the encoder projector.",
+    )
+    parser.add_argument(
+        "--use-flash-attn",
+        type=str2bool,
+        default=True,
+        help="Whether to use flash attention.",
+    )
+
+    parser.add_argument(
+        "--use-lora",
+        type=str2bool,
+        default=False,
+        help="Whether to use lora to fine-tune llm.",
+    )
+
+    parser.add_argument(
+        "--unfreeze-llm",
+        type=str2bool,
+        default=False,
+        help="Whether to unfreeze llm during training.",
+    )
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--tensorboard",
+        type=str2bool,
+        default=True,
+        help="Should various information be logged in tensorboard.",
+    )
+
+    parser.add_argument(
+        "--num-epochs",
+        type=int,
+        default=10,
+        help="Number of epochs to train.",
+    )
+
+    parser.add_argument(
+        "--start-epoch",
+        type=int,
+        default=1,
+        help="""Resume training from this epoch. It should be positive.
+        If larger than 1, it will load checkpoint from
+        exp-dir/epoch-{start_epoch-1}.pt
+        """,
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="whisper_qwen/exp",
+        help="""The experiment dir.
+        It specifies the directory where all training related
+        files, e.g., checkpoints, log, etc, are saved
+        """,
+    )
+
+    parser.add_argument(
+        "--pretrained-model-path",
+        type=str,
+        default=None,
+        help="""The path to the pretrained model if it is not None. Training will
+        start from this model. e.g. ./wenetspeech/ASR/whisper/exp_large_v2/epoch-4-avg-3.pt
+        """,
+    )
+
+    parser.add_argument(
+        "--sampler-state-dict-path",
+        type=str,
+        default=None,
+        help="""The path to the sampler state dict if it is not None. Training will start from this sampler state dict.
+        """,
+    )
+
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=42,
+        help="The seed for random generators intended for reproducibility",
+    )
+
+    parser.add_argument(
+        "--use-fp16",
+        type=str2bool,
+        default=True,
+        help="Whether to use half precision training.",
+    )
+
+    parser.add_argument(
+        "--use-aishell",
+        type=str2bool,
+        default=True,
+        help="Whether to only use aishell1 dataset for training.",
+    )
+
+    parser = deepspeed.add_config_arguments(parser)
+    add_model_arguments(parser)
+
+    return parser
+
+
+def get_params() -> AttributeDict:
+    """Return a dict containing training parameters.
+
+    All training related parameters that are not passed from the commandline
+    are saved in the variable `params`.
+
+    Commandline options are merged into `params` after they are parsed, so
+    you can also access them via `params`.
+
+    Explanation of options saved in `params`:
+
+        - frame_shift_ms: The frame shift in milliseconds.
+        - allowed_excess_duration_ratio: The allowed excess duration ratio.
+        - best_train_loss: The best training loss so far.
+        - best_valid_loss: The best validation loss so far.
+        - best_train_epoch: The epoch where the best training loss is achieved.
+        - best_valid_epoch: The epoch where the best validation loss is achieved.
+        - batch_idx_train: The batch index of the current batch.
+        - log_interval: Log training stats every `log_interval` batches.
+        - reset_interval: Reset the stats every `reset_interval` batches.
+        - valid_interval: Run validation every `valid_interval` batches.
+        - env_info: The environment information.
+    """
+    params = AttributeDict(
+        {
+            "allowed_excess_duration_ratio": 0.1,
+            "subsampling_factor": 2,
+            "frame_shift_ms": 10,
+            "best_train_loss": float("inf"),
+            "best_valid_loss": float("inf"),
+            "best_train_epoch": -1,
+            "best_valid_epoch": -1,
+            "batch_idx_train": 0,
+            "log_interval": 50,
+            "reset_interval": 200,
+            "valid_interval": 5000,
+            "env_info": get_env_info(),
+        }
+    )
+
+    return params
+
+
+def compute_loss(
+    params: AttributeDict,
+    tokenizer: AutoTokenizer,
+    model: nn.Module,
+    batch: dict,
+    is_training: bool,
+) -> Tuple[Tensor, MetricsTracker]:
+    """
+    Compute the loss for the given batch.
+    Args:
+        params:
+            It is returned by :func:`get_params`.
+        tokenizer:
+            The tokenizer used to encode the text.
+        model:
+            The model for training.
+        batch:
+            A batch of data. See `lhotse.dataset.K2SpeechRecognitionDataset()`
+            for the content in it.
+        is_training:
+            Whether it is training.
+    Returns:
+        Return a tuple of two elements. The first element is the loss tensor.
+    """
+
+    def preprocess(
+        messages,
+        tokenizer: transformers.PreTrainedTokenizer,
+        max_len: int,
+    ) -> Dict:
+        """Preprocesses the data for supervised fine-tuning."""
+        texts = []
+        TEMPLATE = "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content']}}{% if loop.last %}{{ '<|im_end|>'}}{% else %}{{ '<|im_end|>\n' }}{% endif %}{% endfor %}"
+        for i, msg in enumerate(messages):
+            texts.append(
+                tokenizer.apply_chat_template(
+                    msg,
+                    tokenize=True,
+                    chat_template=TEMPLATE,
+                    add_generation_prompt=False,
+                    padding="longest",  # FIX me change padding to longest
+                    max_length=max_len,
+                    truncation=True,
+                )
+            )
+        # padding texts to the same length, texts is a list of list, padding with tokenzier.pad_token_id
+        max_len_texts = max([len(text) for text in texts])
+        if tokenizer.padding_side == "right":
+            texts = [
+                text + [tokenizer.pad_token_id] * (max_len_texts - len(text))
+                for text in texts
+            ]
+        else:
+            texts = [
+                [tokenizer.pad_token_id] * (max_len_texts - len(text)) + text
+                for text in texts
+            ]
+        input_ids = torch.tensor(texts, dtype=torch.int)
+        # response = tokenizer.batch_decode(input_ids, skip_special_tokens=True)[0]
+        target_ids = input_ids.clone()
+        target_ids[target_ids == tokenizer.pad_token_id] = IGNORE_TOKEN_ID
+        # mask all tokens before token_id 151646 with IGNORE_TOKEN_ID
+        # first get the indices of the tokens
+        mask_prompt = True
+        if mask_prompt:
+            mask_indices = torch.where(
+                input_ids == tokenizer.convert_tokens_to_ids("assistant")
+            )
+            for i in range(mask_indices[0].size(0)):
+                row = mask_indices[0][i]
+                col = mask_indices[1][i]
+                # + 2 to  skip: 'assistant', '\n'
+                target_ids[row, : col + 2] = IGNORE_TOKEN_ID
+
+        attention_mask = input_ids.ne(tokenizer.pad_token_id)
+
+        return input_ids, attention_mask, target_ids
+
+    device = next(model.parameters()).device
+    feature = batch["inputs"]
+
+    assert feature.ndim == 3
+    feature = feature.to(device)
+    feature = feature.transpose(1, 2)  # (N, C, T)
+
+    batch_idx_train = params.batch_idx_train
+    supervisions = batch["supervisions"]
+    texts = batch["supervisions"]["text"]
+
+    messages = []
+    for i, text in enumerate(texts):
+        text = text.replace(" ", "")
+        message = [
+            {"role": "user", "content": f"{DEFAULT_SPEECH_TOKEN}请转写音频为文字"},
+            {"role": "assistant", "content": text},
+        ]
+        messages.append(message)
+
+    input_ids, attention_mask, target_ids = preprocess(messages, tokenizer, max_len=128)
+
+    target_ids = target_ids.type(torch.LongTensor)
+    input_ids = input_ids.type(torch.LongTensor)
+
+    with torch.set_grad_enabled(is_training):
+        model_outputs, acc = model(
+            fbank=feature,
+            input_ids=input_ids.to(device),
+            attention_mask=attention_mask.to(device),
+            labels=target_ids.to(device),
+        )
+        loss = model_outputs.loss
+    assert loss.requires_grad == is_training
+
+    info = MetricsTracker()
+    with warnings.catch_warnings():
+        warnings.simplefilter("ignore")
+        feature_lens = supervisions["num_frames"]
+        info["frames"] = (feature_lens // params.subsampling_factor).sum().item()
+
+    # Note: We use reduction=sum while computing the loss.
+    info["loss"] = loss.detach().cpu().item()
+    info["acc"] = (
+        acc * info["frames"]
+    )  # WAR: to avoid normalization by the number of frames
+
+    return loss, info
+
+
+def compute_validation_loss(
+    params: AttributeDict,
+    tokenizer: whisper.tokenizer.Tokenizer,
+    model: nn.Module,
+    valid_dl: torch.utils.data.DataLoader,
+    world_size: int = 1,
+) -> MetricsTracker:
+    """Run the validation process."""
+    model.eval()
+
+    tot_loss = MetricsTracker()
+
+    for batch_idx, batch in enumerate(valid_dl):
+        with torch.cuda.amp.autocast(enabled=params.use_fp16):
+            loss, loss_info = compute_loss(
+                params=params,
+                tokenizer=tokenizer,
+                model=model,
+                batch=batch,
+                is_training=False,
+            )
+        assert loss.requires_grad is False
+        tot_loss = tot_loss + loss_info
+
+    if world_size > 1:
+        tot_loss.reduce(loss.device)
+
+    loss_value = tot_loss["loss"] / tot_loss["frames"]
+    if loss_value < params.best_valid_loss:
+        params.best_valid_epoch = params.cur_epoch
+        params.best_valid_loss = loss_value
+
+    return tot_loss
+
+
+def train_one_epoch(
+    params: AttributeDict,
+    tokenizer: AutoTokenizer,
+    model: nn.Module,
+    optimizer: torch.optim.Optimizer,
+    scheduler: torch.optim.lr_scheduler,
+    train_dl: torch.utils.data.DataLoader,
+    valid_dl: torch.utils.data.DataLoader,
+    tb_writer: Optional[SummaryWriter] = None,
+    world_size: int = 1,
+    rank: int = 0,
+) -> None:
+    """Train the model for one epoch.
+
+    The training loss from the mean of all frames is saved in
+    `params.train_loss`. It runs the validation process every
+    `params.valid_interval` batches.
+
+    Args:
+      params:
+        It is returned by :func:`get_params`.
+      model:
+        The model for training.
+      optimizer:
+        The optimizer we are using.
+      scheduler:
+        The learning rate scheduler, we call step() every step.
+      train_dl:
+        Dataloader for the training dataset.
+      valid_dl:
+        Dataloader for the validation dataset.
+      scaler:
+        The scaler used for mix precision training.
+      model_avg:
+        The stored model averaged from the start of training.
+      tb_writer:
+        Writer to write log messages to tensorboard.
+      world_size:
+        Number of nodes in DDP training. If it is 1, DDP is disabled.
+      rank:
+        The rank of the node in DDP training. If no DDP is used, it should
+        be set to 0.
+    """
+    model.train()
+    model.encoder.eval()
+    if not params.unfreeze_llm:
+        model.llm.eval()
+
+    tot_loss = MetricsTracker()
+
+    for batch_idx, batch in enumerate(train_dl):
+        params.batch_idx_train += 1
+        batch_size = len(batch["supervisions"]["text"])
+        if batch_idx % params.valid_interval == 0:
+            logging.info("Computing validation loss")
+            valid_info = compute_validation_loss(
+                params=params,
+                tokenizer=tokenizer,
+                model=model,
+                valid_dl=valid_dl,
+                world_size=world_size,
+            )
+            model.train()
+            model.encoder.eval()
+            if not params.unfreeze_llm:
+                model.llm.eval()
+            logging.info(f"Epoch {params.cur_epoch}, validation: {valid_info}")
+            logging.info(
+                f"Maximum memory allocated so far is {torch.cuda.max_memory_allocated()//1000000}MB"
+            )
+            if tb_writer is not None:
+                valid_info.write_summary(
+                    tb_writer, "train/valid_", params.batch_idx_train
+                )
+            if batch_idx != 0:
+                model.save_checkpoint(
+                    save_dir=params.exp_dir,
+                    tag=f"epoch-{params.cur_epoch}-checkpoint-{batch_idx}",
+                    client_state={},
+                    exclude_frozen_parameters=True,
+                )
+
+                if rank == 0:
+                    convert_zero_checkpoint_to_fp32_state_dict(
+                        params.exp_dir,
+                        f"{params.exp_dir}/epoch-{params.cur_epoch}-checkpoint-{batch_idx}.pt",
+                        tag=f"epoch-{params.cur_epoch}-checkpoint-{batch_idx}",
+                        exclude_frozen_parameters=True,
+                    )
+                    # save sampler state dict into checkpoint
+                    sampler_state_dict = train_dl.sampler.state_dict()
+                    torch.save(
+                        sampler_state_dict,
+                        f"{params.exp_dir}/epoch-{params.cur_epoch}-checkpoint-{batch_idx}-sampler.pt",
+                    )
+                    os.system(
+                        f"rm -rf {params.exp_dir}/epoch-{params.cur_epoch}-checkpoint-{batch_idx}"
+                    )
+        try:
+            with torch.cuda.amp.autocast(enabled=params.use_fp16):
+                loss, loss_info = compute_loss(
+                    params=params,
+                    tokenizer=tokenizer,
+                    model=model,
+                    batch=batch,
+                    is_training=True,
+                )
+            # summary stats
+            tot_loss = (tot_loss * (1 - 1 / params.reset_interval)) + loss_info
+
+            # NOTE: We use reduction==sum and loss is computed over utterances
+            # in the batch and there is no normalization to it so far.
+
+            # deepspeed's backward() is different from torch's backward()
+            # in that it does not accept a loss tensor as input.
+            # It computes the loss internally.
+            model.backward(loss)
+            model.step()
+
+        except:  # noqa
+            display_and_save_batch(batch, params=params)
+            raise
+
+        if batch_idx % params.log_interval == 0:
+            try:
+                cur_lr = scheduler.get_last_lr()[0]
+            except:  # noqa
+                cur_lr = 0.0
+
+            logging.info(
+                f"Epoch {params.cur_epoch}, "
+                f"batch {batch_idx}, loss[{loss_info}], "
+                f"tot_loss[{tot_loss}], batch size: {batch_size}, "
+                f"lr: {cur_lr:.2e}, "
+            )
+
+            if tb_writer is not None:
+                tb_writer.add_scalar(
+                    "train/learning_rate", cur_lr, params.batch_idx_train
+                )
+
+                loss_info.write_summary(
+                    tb_writer, "train/current_", params.batch_idx_train
+                )
+                tot_loss.write_summary(tb_writer, "train/tot_", params.batch_idx_train)
+
+    loss_value = tot_loss["loss"] / tot_loss["frames"]
+    params.train_loss = loss_value
+    if params.train_loss < params.best_train_loss:
+        params.best_train_epoch = params.cur_epoch
+        params.best_train_loss = params.train_loss
+
+
+def run(rank, world_size, args):
+    """
+    Args:
+      rank:
+        It is a value between 0 and `world_size-1`, which is
+        passed automatically by `mp.spawn()` in :func:`main`.
+        The node with rank 0 is responsible for saving checkpoint.
+      world_size:
+        Number of GPUs for DDP training.
+      args:
+        The return value of get_parser().parse_args()
+    """
+    params = get_params()
+    params.update(vars(args))
+
+    fix_random_seed(params.seed)
+
+    setup_logger(f"{params.exp_dir}/log/log-train")
+    logging.info(params)
+
+    logging.info("About to create model")
+
+    replace_whisper_encoder_forward()
+    whisper_model = whisper.load_model(params.speech_encoder_path_or_name, "cpu")
+    speech_encoder = whisper_model.encoder
+    speech_encoder_dim = whisper_model.dims.n_audio_state
+    for name, param in speech_encoder.named_parameters():
+        param.requires_grad = False
+
+    tokenizer = AutoTokenizer.from_pretrained(params.llm_path_or_name)
+    if params.use_flash_attn:
+        attn_implementation = "flash_attention_2"
+        # torch_dtype=torch.bfloat16 FIX ME
+        torch_dtype = torch.float16
+        tokenizer.padding_side = "left"
+
+    else:
+        attn_implementation = "eager"
+        torch_dtype = torch.float16
+        tokenizer.padding_side = "right"
+
+    llm = AutoModelForCausalLM.from_pretrained(
+        params.llm_path_or_name,
+        attn_implementation=attn_implementation,
+        torch_dtype=torch_dtype,
+    )
+
+    if not params.unfreeze_llm:
+        for name, param in llm.named_parameters():
+            param.requires_grad = False
+    else:
+        if params.use_lora:
+            lora_config = LoraConfig(
+                r=64,
+                lora_alpha=16,
+                target_modules=[
+                    "q_proj",
+                    "k_proj",
+                    "v_proj",
+                    "o_proj",
+                    "up_proj",
+                    "gate_proj",
+                    "down_proj",
+                ],
+                lora_dropout=0.05,
+                task_type="CAUSAL_LM",
+            )
+            llm = get_peft_model(llm, lora_config)
+            llm.print_trainable_parameters()
+
+    special_tokens_dict = {"additional_special_tokens": [DEFAULT_SPEECH_TOKEN]}
+    tokenizer.add_special_tokens(special_tokens_dict)
+    llm.config.pad_token_id = tokenizer.pad_token_id
+    llm.config.default_speech_token_id = tokenizer.convert_tokens_to_ids(
+        DEFAULT_SPEECH_TOKEN
+    )
+
+    encoder_projector = EncoderProjector(
+        speech_encoder_dim, llm.config.hidden_size, params.encoder_projector_ds_rate
+    )
+
+    model = SPEECH_LLM(
+        speech_encoder,
+        llm,
+        encoder_projector,
+    )
+
+    if params.pretrained_model_path:
+        checkpoint = torch.load(params.pretrained_model_path, map_location="cpu")
+        missing_keys, unexpected_keys = model.load_state_dict(checkpoint, strict=False)
+
+    num_param = sum([p.numel() for p in model.parameters()])
+    logging.info(f"Number of model parameters: {num_param}")
+
+    logging.info("Trainable parameters (excluding model.eval modules):")
+    for name, param in model.named_parameters():
+        if param.requires_grad:
+            logging.info(f"{name}: {param.shape}")
+
+    if torch.cuda.is_available():
+        device = torch.device("cuda", rank)
+    else:
+        device = torch.device("cpu")
+    logging.info(f"Device: {device}")
+    model.to(device)
+
+    assert params.deepspeed
+    logging.info("Using DeepSpeed")
+    model, optimizer, _, scheduler = deepspeed.initialize(
+        args=params, model=model, model_parameters=model.parameters()
+    )
+
+    data_module = AsrDataModule(args)
+    multi_dataset = MultiDataset(args.manifest_dir)
+
+    def remove_short_and_long_utt(c: Cut):
+        # Keep only utterances with duration between 1 second and 20 seconds
+        #
+        # Caution: There is a reason to select 20.0 here. Please see
+        # ../local/display_manifest_statistics.py
+        #
+        # You should use ../local/display_manifest_statistics.py to get
+        # an utterance duration distribution for your dataset to select
+        # the threshold
+        if c.duration < 1.0 or c.duration > 20.0:
+            # logging.warning(
+            #    f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
+            # )
+            return False
+        return True
+
+    if params.use_aishell:
+        train_cuts = multi_dataset.aishell_train_cuts()
+    else:
+        train_cuts = multi_dataset.train_cuts()
+
+    train_cuts = train_cuts.filter(remove_short_and_long_utt)
+
+    sampler_state_dict = None
+    if params.sampler_state_dict_path:
+        sampler_state_dict = torch.load(params.sampler_state_dict_path)
+        sampler_state_dict["max_duration"] = params.max_duration
+
+    train_dl = data_module.train_dataloaders(
+        train_cuts, sampler_state_dict=sampler_state_dict
+    )
+
+    if params.use_aishell:
+        valid_cuts = multi_dataset.aishell_dev_cuts()
+    else:
+        valid_cuts = multi_dataset.dev_cuts()
+    valid_dl = data_module.valid_dataloaders(valid_cuts)
+
+    if args.tensorboard and rank == 0:
+        tb_writer = SummaryWriter(log_dir=f"{params.exp_dir}/tensorboard")
+    else:
+        tb_writer = None
+
+    logging.info(f"start training from epoch {params.start_epoch}")
+    for epoch in range(params.start_epoch, params.num_epochs + 1):
+
+        fix_random_seed(params.seed + epoch - 1)
+        train_dl.sampler.set_epoch(epoch - 1)
+
+        if tb_writer is not None:
+            tb_writer.add_scalar("train/epoch", epoch, params.batch_idx_train)
+
+        params.cur_epoch = epoch
+
+        train_one_epoch(
+            params=params,
+            tokenizer=tokenizer,
+            model=model,
+            optimizer=optimizer,
+            scheduler=scheduler,
+            train_dl=train_dl,
+            valid_dl=valid_dl,
+            tb_writer=tb_writer,
+            world_size=world_size,
+            rank=rank,
+        )
+
+        model.save_checkpoint(
+            save_dir=params.exp_dir,
+            tag=f"zero-epoch-{params.cur_epoch}",
+            client_state={},
+            exclude_frozen_parameters=True,
+        )
+        if rank == 0:
+            convert_zero_checkpoint_to_fp32_state_dict(
+                params.exp_dir,
+                f"{params.exp_dir}/epoch-{params.cur_epoch}",
+                tag=f"zero-epoch-{params.cur_epoch}",
+                exclude_frozen_parameters=True,
+            )
+            # save sampler state dict into checkpoint
+            sampler_state_dict = train_dl.sampler.state_dict()
+            torch.save(
+                sampler_state_dict,
+                f"{params.exp_dir}/epoch-{params.cur_epoch}-sampler.pt",
+            )
+
+            os.system(f"rm -rf {params.exp_dir}/zero-epoch-{params.cur_epoch}")
+
+    logging.info("Done!")
+
+
+def display_and_save_batch(
+    batch: dict,
+    params: AttributeDict,
+) -> None:
+    """Display the batch statistics and save the batch into disk.
+
+    Args:
+      batch:
+        A batch of data. See `lhotse.dataset.K2SpeechRecognitionDataset()`
+        for the content in it.
+      params:
+        Parameters for training. See :func:`get_params`.
+    """
+    from lhotse.utils import uuid4
+
+    filename = f"{params.exp_dir}/batch-{uuid4()}.pt"
+    logging.info(f"Saving batch to {filename}")
+    torch.save(batch, filename)
+
+    supervisions = batch["supervisions"]
+    features = batch["inputs"]
+
+    logging.info(f"features shape: {features.shape}")
+
+
+def main():
+    parser = get_parser()
+    AsrDataModule.add_arguments(parser)
+    args = parser.parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    world_size = get_world_size()
+    rank = get_rank()
+
+    torch.set_num_threads(1)
+    torch.set_num_interop_threads(1)
+    warnings.filterwarnings("ignore", category=FutureWarning)
+    run(rank=rank, world_size=world_size, args=args)
+
+
+if __name__ == "__main__":
+    main()