add speech continuation pretraining

egs/speech_llm/SPEECH2SPEECH/prepare.sh

@@ -122,7 +122,7 @@ if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
 fi
 
 if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
-  log "stage 1: Compute fbank feature from huggingface"
+  log "stage 6: Compute fbank feature from huggingface"
   # CUDA_VISIBLE_DEVICES=0 python3 local/compute_whisper_fbank.py \
   #  --num-mel-bins 80 --whisper-fbank True --resample-to-16kHz True --speed-perturb False \
   #  --out-dir data/fbank_voice_assistant \
@@ -161,10 +161,7 @@ if [ $stage -le 8 ] && [ $stop_stage -ge 8 ]; then
    --subset test --split test \
    --audio-key audio --text-key text \
    --prefix gigaspeech
-fi
 
-if [ $stage -le 9 ] && [ $stop_stage -ge 9 ]; then
-  log "stage 9: Compute fbank feature from huggingface"
   CUDA_VISIBLE_DEVICES=0 python3 local/compute_whisper_fbank.py \
    --num-mel-bins 80 --whisper-fbank True --resample-to-16kHz True --speed-perturb True \
    --out-dir data/fbank_gigaspeech \
@@ -195,7 +192,7 @@ fi
 
 
 if [ $stage -le 11 ] && [ $stop_stage -ge 11 ]; then
-  log "stage 11: Decoding EN, only support batch_size=1 for now."
+  log "stage 11: Decoding EN, val set only support batch_size=1 for now."
   exp_dir=./qwen_omni/exp_speech2speech_en_continue
   # cd $exp_dir && ln -s ../../models/qwen-omni-like-speech2speech-belle-1.4M/pytorch_model.bin epoch-999.pt && cd -
   python3 ./qwen_omni/decode.py \
@@ -256,3 +253,71 @@ if [ $stage -le 14 ] && [ $stop_stage -ge 14 ]; then
       --output-dir test_result
   done
 fi
+
+
+if [ $stage -le 15 ] && [ $stop_stage -ge 15 ]; then
+  log "stage 15: Training Speech2Speech Model, adaptor only"
+  exp_dir=./qwen_omni/exp_speech2text
+  ngpu=2
+  torchrun --nproc_per_node $ngpu ./qwen_omni/train.py \
+    --max-duration 600 \
+    --enable-musan False \
+    --audio-key audio --text-key continuation \
+    --exp-dir $exp_dir \
+    --speech-encoder-path-or-name models/large-v2.pt \
+    --llm-path-or-name Qwen/Qwen2.5-0.5B-Instruct \
+    --on-the-fly-feats True \
+    --deepspeed \
+    --deepspeed_config ./qwen_omni/ds_config_zero1.json \
+    --use-flash-attn True \
+    --dataset-format speech_continuation \
+    --start-epoch 2 --pretrained-model-path $exp_dir/epoch-1/pytorch_model.bin \
+    --use-lora False --unfreeze-llm False --unfreeze-speech-projector True --enable-speech-output False
+fi
+
+if [ $stage -le 16 ] && [ $stop_stage -ge 16 ]; then
+  log "stage 16: Training Speech2Speech Model, adaptor only"
+  exp_dir=./qwen_omni/exp_speech2text
+  ngpu=4
+
+  latest_checkpoint_step=-1
+  # Check if exp_dir exists and is a directory
+  if [ -d "$exp_dir" ]; then
+    # List directories matching checkpoint-* and find the one with the largest step number
+    for checkpoint_dir in $(ls -d $exp_dir/checkpoint-*/ 2>/dev/null | sort -V); do
+      checkpoint_name=$(basename "$checkpoint_dir") # e.g., checkpoint-1000
+      # Extract step number using parameter expansion
+      current_step=${checkpoint_name#checkpoint-}
+      # Ensure current_step is a number
+      if [[ "$current_step" =~ ^[0-9]+$ ]] && [ "$current_step" -gt "$latest_checkpoint_step" ]; then
+        latest_checkpoint_step=$current_step
+      fi
+    done
+  fi
+
+  train_cmd_args="--max-duration 1200 \
+    --enable-musan False \
+    --audio-key audio --text-key continuation \
+    --exp-dir $exp_dir \
+    --speech-encoder-path-or-name models/large-v2.pt \
+    --llm-path-or-name Qwen/Qwen2.5-0.5B-Instruct \
+    --on-the-fly-feats True \
+    --deepspeed \
+    --huggingface-dataset-path-or-name /lustre/fsw/general_sa/yuekaiz/s2s \
+    --deepspeed_config ./qwen_omni/ds_config_zero1.json \
+    --use-flash-attn True \
+    --dataset-format speech_continuation \
+    --use-lora False --unfreeze-llm False --unfreeze-speech-projector True --enable-speech-output False"
+
+  if [ "$latest_checkpoint_step" -ge 0 ]; then
+    log "Continuing training from checkpoint-$latest_checkpoint_step"
+    step=$latest_checkpoint_step
+    train_cmd_args="$train_cmd_args --pretrained-model-path $exp_dir/checkpoint-${step}/pytorch_model.bin --sampler-state-dict-path $exp_dir/checkpoint-${step}/sampler.pt"
+  else
+    log "Starting training from scratch as no checkpoint was found in $exp_dir"
+    # No pretrained model or sampler state dict needed for the first run
+  fi
+
+  torchrun --nproc_per_node $ngpu ./qwen_omni/train.py \
+    $train_cmd_args
+fi

egs/speech_llm/SPEECH2SPEECH/qwen_omni/data_module.py

@@ -24,7 +24,7 @@ from pathlib import Path
 from typing import Any, Dict, Optional
 
 import torch
-from datasets import load_dataset
+from datasets import interleave_datasets, load_dataset
 from lhotse import (
     CutSet,
     WhisperFbank,
@@ -36,6 +36,7 @@ from lhotse.dataset import (  # noqa F401 for PrecomputedFeatures
     CutConcatenate,
     CutMix,
     DynamicBucketingSampler,
+    PerturbSpeed,
     PrecomputedFeatures,
     SimpleCutSampler,
     SpecAugment,
@@ -47,7 +48,7 @@ from lhotse.dataset.input_strategies import (  # noqa F401 For AudioSamples
 from lhotse.utils import fix_random_seed
 from speech_dataset import K2SpeechRecognitionDataset
 from torch.utils.data import DataLoader
-from utils import str2bool
+from utils import get_rank, str2bool
 
 
 class _SeedWorkers:
@@ -123,6 +124,14 @@ class AsrDataModule:
             "extraction. Will drop existing precomputed feature manifests "
             "if available.",
         )
+        group.add_argument(
+            "--on-the-fly-speed-perturb",
+            type=str2bool,
+            default=True,
+            help="When enabled, use on-the-fly speed perturbation. "
+            "Will drop existing precomputed feature manifests "
+            "if available.",
+        )
         group.add_argument(
             "--shuffle",
             type=str2bool,
@@ -188,27 +197,27 @@ class AsrDataModule:
         group.add_argument(
             "--huggingface-dataset-path-or-name",
             type=str,
-            default="/workspace/Belle_1.4M-SLAM-Omni",
+            default=None,
             help="The path or name of the Huggingface dataset",
         )
         group.add_argument(
             "--audio-key",
             type=str,
-            default="question_audio",
+            default="audio",
             help="The key in the Huggingface dataset containing the audio data",
         )
         group.add_argument(
             "--text-key",
             type=str,
-            default="answer",
+            default="text",
             help="The key in the Huggingface dataset containing the text data",
         )
-        group.add_argument(
+        # group.add_argument(
-            "--resample-to-16kHz",
+        #     "--resample-to-16kHz",
-            type=str2bool,
+        #     type=str2bool,
-            default=True,
+        #     default=True,
-            help="Resample audio to 16kHz. Default: False.",
+        #     help="Resample audio to 16kHz. Default: False.",
-        )
+        # )
 
     def train_dataloaders(
         self,
@@ -232,6 +241,8 @@ class AsrDataModule:
             )
         else:
             logging.info("Disable MUSAN")
+        if self.args.on_the_fly_speed_perturb and self.args.on_the_fly_feats:
+            transforms = [PerturbSpeed(factors=[0.9, 1.1], p=2 / 3)] + transforms
 
         input_transforms = []
         if self.args.enable_spec_aug:
@@ -260,9 +271,11 @@ class AsrDataModule:
             logging.info("Disable SpecAugment")
 
         logging.info("About to create train dataset")
+        rank = get_rank()
+
         train = K2SpeechRecognitionDataset(
             input_strategy=OnTheFlyFeatures(
-                WhisperFbank(WhisperFbankConfig(num_filters=80, device="cuda"))
+                WhisperFbank(WhisperFbankConfig(num_filters=80, device=f"cuda:{rank}"))
             )
             if self.args.on_the_fly_feats
             else eval(self.args.input_strategy)(),
@@ -271,26 +284,6 @@ class AsrDataModule:
             return_cuts=self.args.return_cuts,
         )
 
-        # if self.args.on_the_fly_feats:
-        #     # NOTE: the PerturbSpeed transform should be added only if we
-        #     # remove it from data prep stage.
-        #     # Add on-the-fly speed perturbation; since originally it would
-        #     # have increased epoch size by 3, we will apply prob 2/3 and use
-        #     # 3x more epochs.
-        #     # Speed perturbation probably should come first before
-        #     # concatenation, but in principle the transforms order doesn't have
-        #     # to be strict (e.g. could be randomized)
-        #     # transforms = [PerturbSpeed(factors=[0.9, 1.1], p=2/3)] + transforms   # noqa
-        #     # Drop feats to be on the safe side.
-        #     train = K2SpeechRecognitionDataset(
-        #         cut_transforms=transforms,
-        #         input_strategy=OnTheFlyFeatures(
-        #             WhisperFbank(WhisperFbankConfig(num_filters=80, device="cuda"))
-        #         ),
-        #         input_transforms=input_transforms,
-        #         return_cuts=self.args.return_cuts,
-        #     )
-
         if self.args.bucketing_sampler:
             logging.info("Using DynamicBucketingSampler.")
             train_sampler = DynamicBucketingSampler(
@@ -298,8 +291,7 @@ class AsrDataModule:
                 max_duration=self.args.max_duration,
                 shuffle=self.args.shuffle,
                 num_buckets=self.args.num_buckets,
-                buffer_size=self.args.num_buckets * 2000,
+                buffer_size=self.args.num_buckets * 1000,
-                shuffle_buffer_size=self.args.num_buckets * 5000,
                 drop_last=self.args.drop_last,
             )
         else:
@@ -339,10 +331,10 @@ class AsrDataModule:
                 CutSet for validation.
         """
         logging.info("About to create dev dataset")
-
+        rank = get_rank()
         validate = K2SpeechRecognitionDataset(
             input_strategy=OnTheFlyFeatures(
-                WhisperFbank(WhisperFbankConfig(num_filters=80, device="cuda"))
+                WhisperFbank(WhisperFbankConfig(num_filters=80, device=f"cuda:{rank}"))
             )
             if self.args.on_the_fly_feats
             else eval(self.args.input_strategy)(),
@@ -470,25 +462,231 @@ class AsrDataModule:
         )
         return {"test": VoiceAssistant_cuts}
 
-    # def train_cuts_en_vocalnet(self) -> CutSet:
+    @lru_cache()
+    def train_cuts_ultravox(self) -> CutSet:
+        logging.info("About to get train cuts")
+        if self.args.huggingface_dataset_path_or_name is not None:
+            librispeech_path = (
+                self.args.huggingface_dataset_path_or_name + "/librispeech_asr"
+            )
+            people_speech_path = (
+                self.args.huggingface_dataset_path_or_name + "/peoples_speech"
+            )
+            gigaspeech_path = self.args.huggingface_dataset_path_or_name + "/gigaspeech"
+        else:
+            librispeech_path = "fixie-ai/librispeech_asr"
+            people_speech_path = "fixie-ai/peoples_speech"
+            gigaspeech_path = "fixie-ai/gigaspeech"
+        # 148_688
+        librispeech_other = load_dataset(
+            librispeech_path, "other", split="train.500", streaming=True
+        )
+        # 104_014
+        librispeech_clean_360 = load_dataset(
+            librispeech_path, "clean", split="train.360", streaming=True
+        )
+        # 28_539
+        librispeech_clean_100 = load_dataset(
+            librispeech_path, "clean", split="train.100", streaming=True
+        )
+
+        # 1_501_271
+        people_speech_clean = load_dataset(
+            people_speech_path, "clean", split="train", streaming=True
+        )
+        # 548_000
+        people_speech_dirty_sa = load_dataset(
+            people_speech_path, "dirty_sa", split="train", streaming=True
+        )
+
+        # 8_266_422
+
+        gigaspeech = load_dataset(
+            gigaspeech_path, "xl-empty-audio-removed", split="train", streaming=True
+        )
+
+        librispeech_clean_100_cuts = CutSet.from_huggingface_dataset(
+            librispeech_clean_100,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+
+        librispeech_other_cuts = CutSet.from_huggingface_dataset(
+            librispeech_other,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+
+        librispeech_clean_360_cuts = CutSet.from_huggingface_dataset(
+            librispeech_clean_360,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+
+        gigaspeech_cuts = CutSet.from_huggingface_dataset(
+            gigaspeech, audio_key=self.args.audio_key, text_key=self.args.text_key
+        )
+
+        people_speech_clean_cuts = CutSet.from_huggingface_dataset(
+            people_speech_clean,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+
+        people_speech_dirty_sa_cuts = CutSet.from_huggingface_dataset(
+            people_speech_dirty_sa,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+
+        return CutSet.mux(
+            librispeech_clean_100_cuts,
+            librispeech_clean_360_cuts,
+            librispeech_other_cuts,
+            gigaspeech_cuts,
+            people_speech_clean_cuts,
+            people_speech_dirty_sa_cuts,
+            weights=[
+                28539,
+                104014,
+                148688,
+                8266422,
+                1501271,
+                548000,
+            ],
+        )
+
+    # @lru_cache()
+    # def train_cuts_ultravox(self) -> CutSet:
     #     logging.info("About to get train cuts")
-    #     VoiceAssistant_cuts = load_manifest_lazy(
+    #     keep_columns = ["audio", "text", "continuation", "id"]
-    #         self.args.manifest_dir / "cuts_debug.jsonl.gz"
+    #     librispeech_path="fixie-ai/librispeech_asr"
-    #     )
+    #     # 148_688
-    #     return VoiceAssistant_cuts
+    #     librispeech_other = load_dataset(librispeech_path, 'other', split='train.500', streaming=True)
+    #     # 104_014
+    #     librispeech_clean_360 = load_dataset(librispeech_path, 'clean', split='train.360', streaming=True)
+    #     # 28_539
+    #     librispeech_clean_100 = load_dataset(librispeech_path, 'clean', split='train.100', streaming=True)
 
-    # @lru_cache()
+    #     cols_to_remove = librispeech_clean_100.column_names
-    # def valid_cuts_en_vocalnet(self) -> CutSet:
+    #     cols_to_remove = [col for col in cols_to_remove if col not in keep_columns]
-    #     logging.info("About to get valid cuts")
+    #     librispeech_clean_100 = librispeech_clean_100.remove_columns(cols_to_remove)
-    #     VoiceAssistant_cuts = load_manifest_lazy(
+    #     librispeech_clean_360 = librispeech_clean_360.remove_columns(cols_to_remove)
-    #         self.args.manifest_dir / "cuts_debug.jsonl.gz"
+    #     librispeech_other = librispeech_other.remove_columns(cols_to_remove)
-    #     )
+    #     people_speech_path="fixie-ai/peoples_speech"
-    #     return VoiceAssistant_cuts
+    #     # 1_501_271
+    #     people_speech_clean = load_dataset(people_speech_path, 'clean', split='train', streaming=True)
+    #     # 548_000
+    #     people_speech_dirty_sa = load_dataset(people_speech_path, 'dirty_sa', split='train', streaming=True)
+    #     cols_to_remove = people_speech_clean.column_names
+    #     cols_to_remove = [col for col in cols_to_remove if col not in keep_columns]
+    #     people_speech_clean = people_speech_clean.remove_columns(cols_to_remove)
+    #     people_speech_dirty_sa = people_speech_dirty_sa.remove_columns(cols_to_remove)
 
-    # @lru_cache()
+    #     # 8_266_422
-    # def test_cuts_en_vocalnet(self) -> CutSet:
+    #     gigaspeech_path="fixie-ai/gigaspeech"
-    #     logging.info("About to get test cuts")
+    #     gigaspeech = load_dataset(gigaspeech_path, 'xl-empty-audio-removed', split='train', streaming=True)
-    #     VoiceAssistant_cuts = load_manifest_lazy(
+    #     # first rename segment_id to id
-    #         self.args.manifest_dir / "cuts_debug.jsonl.gz"
+    #     gigaspeech = gigaspeech.rename_column("segment_id", "id")
+    #     cols_to_remove = gigaspeech.column_names
+    #     cols_to_remove = [col for col in cols_to_remove if col not in keep_columns]
+    #     gigaspeech = gigaspeech.remove_columns(cols_to_remove)
+
+    #     total_item = 104014 + 28539 + 8266422 + 1501271 + 548000 + 148688
+    #     final_datasets = interleave_datasets([
+    #         librispeech_clean_100,
+    #         librispeech_clean_360,
+    #         gigaspeech,
+    #         people_speech_clean,
+    #         people_speech_dirty_sa,
+    #         librispeech_other,
+    #     ], probabilities=[
+    #         28539 / total_item,
+    #         104014 / total_item,
+    #         8266422 / total_item,
+    #         1501271 / total_item,
+    #         548000 / total_item,
+    #         148688 / total_item,
+    #     ])
+
+    #     train_cuts = CutSet.from_huggingface_dataset(
+    #         final_datasets, audio_key=self.args.audio_key, text_key=self.args.text_key
     #     )
-    #     return VoiceAssistant_cuts
+
+    #     return train_cuts
+
+    @lru_cache()
+    def valid_cuts_ultravox(self) -> CutSet:
+        logging.info("About to get valid cuts")
+        librispeech_path = "fixie-ai/librispeech_asr"
+        librispeech_clean_valid = load_dataset(
+            librispeech_path, "clean", split="validation", streaming=True
+        )
+        librispeech_clean_valid_cuts = CutSet.from_huggingface_dataset(
+            librispeech_clean_valid,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+        return librispeech_clean_valid_cuts
+
+    @lru_cache()
+    def train_cuts_librispeech(self) -> CutSet:
+        logging.info("About to get train cuts")
+
+        # librispeech_path="fixie-ai/librispeech_asr"
+        librispeech_path = "/workspace/slam/librispeech_asr"
+        # 148_688
+        librispeech_other = load_dataset(
+            librispeech_path, "other", split="train.500", streaming=True
+        )
+        # 104_014
+        librispeech_clean_360 = load_dataset(
+            librispeech_path, "clean", split="train.360", streaming=True
+        )
+        # 28_539
+        librispeech_clean_100 = load_dataset(
+            librispeech_path, "clean", split="train.100", streaming=True
+        )
+
+        librispeech_clean_100_cuts = CutSet.from_huggingface_dataset(
+            librispeech_clean_100,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+
+        librispeech_other_cuts = CutSet.from_huggingface_dataset(
+            librispeech_other,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+
+        librispeech_clean_360_cuts = CutSet.from_huggingface_dataset(
+            librispeech_clean_360,
+            audio_key=self.args.audio_key,
+            text_key=self.args.text_key,
+        )
+
+        return CutSet.mux(
+            librispeech_clean_100_cuts,
+            librispeech_clean_360_cuts,
+            librispeech_other_cuts,
+            weights=[
+                28539,
+                104014,
+                148688,
+            ],
+        )
+
+    @lru_cache()
+    def train_cuts_gigaspeech(self) -> CutSet:
+        logging.info("About to get train cuts")
+        gigaspeech_path = "fixie-ai/gigaspeech"
+        gigaspeech = load_dataset(
+            gigaspeech_path, "xl-empty-audio-removed", split="train", streaming=True
+        )
+
+        gigaspeech_cuts = CutSet.from_huggingface_dataset(
+            gigaspeech, audio_key=self.args.audio_key, text_key=self.args.text_key
+        )
+
+        return gigaspeech_cuts

egs/speech_llm/SPEECH2SPEECH/qwen_omni/requirements.txt

@@ -10,3 +10,4 @@ transformers>=4.37.0
 flash-attn
 peft
 torchmetrics
+triton==3.3.0 # may be violate with openai-whisper

egs/speech_llm/SPEECH2SPEECH/qwen_omni/train.py

@@ -68,24 +68,26 @@ from transformers import (
     Qwen2Config,
     Qwen2ForCausalLM,
 )
-from whisper_encoder_forward_monkey_patch import replace_whisper_encoder_forward
 
-# from icefall import diagnostics
-from utils import get_rank, get_world_size
 # from icefall.env import get_env_info
+# from icefall import diagnostics
 from utils import (  # filter_uneven_sized_batch,
     AttributeDict,
     MetricsTracker,
+    get_rank,
+    get_world_size,
     setup_logger,
     str2bool,
 )
+from whisper_encoder_forward_monkey_patch import replace_whisper_encoder_forward
 
 DEFAULT_SPEECH_TOKEN = "<speech>"
 try:
-    torch.multiprocessing.set_start_method('spawn')
+    torch.multiprocessing.set_start_method("spawn")
 except RuntimeError:
     pass
 
+
 def set_batch_count(model: nn.Module, batch_count: float) -> None:
     for module in model.modules():
         if hasattr(module, "batch_count"):
@@ -272,7 +274,7 @@ def get_params() -> AttributeDict:
             "batch_idx_train": 0,
             "log_interval": 50,
             "reset_interval": 200,
-            "valid_interval": 5000,
+            "valid_interval": 3000,
             # "env_info": get_env_info(),
         }
     )
@@ -332,6 +334,21 @@ def process_batch_vocalnet(batch: dict):
     return messages, answer_cosyvoice_speech_token
 
 
+def process_batch_speech_continuation(batch: dict):
+    messages = []
+    for i in range(len(batch["supervisions"]["text"])):
+        message = [
+            {
+                "role": "user",
+                "content": f"Continue the following text using less than 50 words:\n\n{DEFAULT_SPEECH_TOKEN}",
+            },
+            {"role": "assistant", "content": batch["supervisions"]["text"][i]},
+        ]
+        # transcript = batch["supervisions"]["cut"][i].custom["text"]
+        messages.append(message)
+    return messages
+
+
 def compute_loss(
     params: AttributeDict,
     tokenizer: AutoTokenizer,
@@ -429,13 +446,13 @@ def compute_loss(
     feature = feature.to(device)
     feature = feature.transpose(1, 2)  # (N, C, T)
 
-    batch_idx_train = params.batch_idx_train
-
     # WAR: TODO FIXME merge process_batch_slam_omni and process_batch_vocalnet
     if params.dataset_format == "slam_omni":
         messages, answer_cosyvoice_speech_token = process_batch_slam_omni(batch)
     elif params.dataset_format == "vocalnet":
         messages, answer_cosyvoice_speech_token = process_batch_vocalnet(batch)
+    elif params.dataset_format == "speech_continuation":
+        messages = process_batch_speech_continuation(batch)
     else:
         raise ValueError(f"Unknown dataset format: {params.dataset_format}")
 
@@ -566,8 +583,11 @@ def train_one_epoch(
         The rank of the node in DDP training. If no DDP is used, it should
         be set to 0.
     """
-    model.encoder_projector.train()
+    # model.encoder_projector.train()
-
+    model.train()
+    model.encoder.eval()
+    if not params.unfreeze_llm:
+        model.llm.eval()
     tot_loss = MetricsTracker()
 
     for batch_idx, batch in enumerate(train_dl):
@@ -583,6 +603,9 @@ def train_one_epoch(
                 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"
@@ -594,7 +617,7 @@ def train_one_epoch(
             if batch_idx != 0:
                 model.save_checkpoint(
                     save_dir=params.exp_dir,
-                    tag=f"epoch-{params.cur_epoch}-checkpoint-{batch_idx}",
+                    tag=f"zero-checkpoint-{params.batch_idx_train}",
                     client_state={},
                     exclude_frozen_parameters=True,
                 )
@@ -602,18 +625,18 @@ def train_one_epoch(
                 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",
+                        f"{params.exp_dir}/checkpoint-{params.batch_idx_train}",
-                        tag=f"epoch-{params.cur_epoch}-checkpoint-{batch_idx}",
+                        tag=f"zero-checkpoint-{params.batch_idx_train}",
                         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",
+                        f"{params.exp_dir}/checkpoint-{params.batch_idx_train}/sampler.pt",
                     )
                     os.system(
-                        f"rm -rf {params.exp_dir}/epoch-{params.cur_epoch}-checkpoint-{batch_idx}"
+                        f"rm -rf {params.exp_dir}/zero-checkpoint-{params.batch_idx_train}"
                     )
         try:
             with torch.amp.autocast("cuda", enabled=params.use_fp16):
@@ -687,9 +710,9 @@ def run(rank, world_size, args):
 
     fix_random_seed(params.seed)
 
+    if rank == 0:
         setup_logger(f"{params.exp_dir}/log/log-train")
     logging.info(params)
-
     logging.info("About to create model")
 
     replace_whisper_encoder_forward()
@@ -698,7 +721,6 @@ def run(rank, world_size, args):
     speech_encoder_dim = whisper_model.dims.n_audio_state
     for name, param in speech_encoder.named_parameters():
         param.requires_grad = False
-    speech_encoder.eval()
 
     tokenizer = AutoTokenizer.from_pretrained(params.llm_path_or_name)
 
@@ -721,7 +743,7 @@ def run(rank, world_size, args):
     if not params.unfreeze_llm:
         for name, param in llm.named_parameters():
             param.requires_grad = False
-        llm.eval()
+
     else:
         if params.use_lora:
             lora_config = LoraConfig(
@@ -809,6 +831,9 @@ def run(rank, world_size, args):
     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)
+        # set params.batch_idx_train according to the checkpoint name
+        if "checkpoint-" in params.pretrained_model_path:
+            params.batch_idx_train = int(params.pretrained_model_path.split("-")[-1])
 
     num_param = sum([p.numel() for p in model.parameters()])
     logging.info(f"Number of model parameters: {num_param}")
@@ -842,11 +867,12 @@ def run(rank, world_size, args):
         # 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 > 30.0:
+        if c.duration < 1.0 or c.duration > 29.5:
-            # logging.warning(
+            logging.warning(
-            #    f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
-            # )
+            )
             return False
+        if "speech_token" in c.custom or "answer_cosyvoice_speech_token" in c.custom:
             codec_len = (
                 len(c.custom["answer_cosyvoice_speech_token"])
                 if "answer_cosyvoice_speech_token" in c.custom
@@ -865,6 +891,11 @@ def run(rank, world_size, args):
     elif params.dataset_format == "vocalnet":
         train_cuts = data_module.train_cuts_en_vocalnet()
         valid_cuts = data_module.valid_cuts_en_vocalnet()
+    elif params.dataset_format == "speech_continuation":
+        # train_cuts = data_module.train_cuts_ultravox()
+        # train_cuts = data_module.train_cuts_gigaspeech()
+        train_cuts = data_module.train_cuts_librispeech()
+        valid_cuts = data_module.valid_cuts_ultravox()
     else:
         raise ValueError(f"Unknown dataset format: {params.dataset_format}")
 
@@ -879,7 +910,7 @@ def run(rank, world_size, args):
     train_dl = data_module.train_dataloaders(
         train_cuts, sampler_state_dict=sampler_state_dict
     )
-
+    # train_dl = data_module.valid_dataloaders(train_cuts)
     valid_dl = data_module.valid_dataloaders(valid_cuts)
 
     if args.tensorboard and rank == 0:
@@ -913,25 +944,25 @@ def run(rank, world_size, args):
 
         model.save_checkpoint(
             save_dir=params.exp_dir,
-            tag=f"epoch-{params.cur_epoch}",
+            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}.pt",
+                f"{params.exp_dir}/epoch-{params.cur_epoch}",
-                tag=f"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",
+                f"{params.exp_dir}/epoch-{params.cur_epoch}/sampler.pt",
             )
 
-            os.system(f"rm -rf {params.exp_dir}/epoch-{params.cur_epoch}")
+            os.system(f"rm -rf {params.exp_dir}/zero-epoch-{params.cur_epoch}")
 
     logging.info("Done!")
 
@@ -971,6 +1002,7 @@ def main():
 
     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)