Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
fix: handle None scheduler_conf in trainer, add scheduler to VAD finetune
游雁b710ffda9146
diff --git a/examples/industrial_data_pretraining/fsmn_vad_streaming/finetune.sh b/examples/industrial_data_pretraining/fsmn_vad_streaming/finetune.shindex 6be48366b..6be4926f9 100755--- a/examples/industrial_data_pretraining/fsmn_vad_streaming/finetune.sh+++ b/examples/industrial_data_pretraining/fsmn_vad_streaming/finetune.sh@@ -61,4 +61,6 @@ torchrun \++train_conf.avg_nbest_model=3 \++train_conf.use_deepspeed=false \++optim_conf.lr=0.00005 \+++scheduler="warmuplr" \+++scheduler_conf.warmup_steps=1000 \++output_dir="${output_dir}" &> ${log_file}diff --git a/funasr/train_utils/trainer_ds.py b/funasr/train_utils/trainer_ds.pyindex f4f319d31..ff649e30e 100644--- a/funasr/train_utils/trainer_ds.py+++ b/funasr/train_utils/trainer_ds.py@@ -1018,7 +1018,7 @@ class Trainer:scheduler = kwargs.get("scheduler", "warmuplr")assert scheduler in scheduler_classesscheduler_class = scheduler_classes.get(scheduler)- scheduler = scheduler_class(optim, **kwargs.get("scheduler_conf"))+ scheduler = scheduler_class(optim, **(kwargs.get("scheduler_conf") or {}))if self.use_deepspeed:import deepspeed@@ -1047,7 +1047,7 @@ class Trainer:Args:opt: TODO."""- return scheduler_class(opt, **kwargs.get("scheduler_conf"))+ return scheduler_class(opt, **(kwargs.get("scheduler_conf") or {}))model, optimizer, _, scheduler = deepspeed.initialize(args=args,