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
auto frontend
游雁524237d7595c
diff --git a/examples/industrial_data_pretraining/llm_asr/demo_train_or_finetune2.sh b/examples/industrial_data_pretraining/llm_asr/demo_train_or_finetune2.shindex b3aac2b70..306e23db9 100644--- a/examples/industrial_data_pretraining/llm_asr/demo_train_or_finetune2.sh+++ b/examples/industrial_data_pretraining/llm_asr/demo_train_or_finetune2.sh@@ -41,6 +41,7 @@ torchrun \++dataset_conf.batch_size=1 \++dataset_conf.num_workers=0 \++train_conf.max_epoch=15 \+++train_conf.save_checkpoint_interval=1000 \++optim_conf.lr=0.0001 \++init_param="${init_param}" \++output_dir="${output_dir}" &> ${log_file} &diff --git a/funasr/train_utils/trainer.py b/funasr/train_utils/trainer.pyindex 50f99f023..60fd96966 100644--- a/funasr/train_utils/trainer.py+++ b/funasr/train_utils/trainer.py@@ -85,7 +85,12 @@ class Trainer:self.batch_total = 0self.use_fp16 = use_fp16self.save_checkpoint_interval = kwargs.get("save_checkpoint_interval", 5000)- self.validate_interval = kwargs.get("validate_interval", 5000)+ self.validate_interval = kwargs.get("validate_interval", -1)+ if self.validate_interval < 0:+ self.validate_interval = self.save_checkpoint_interval+ assert (+ self.save_checkpoint_interval == self.validate_interval+ ), f"save_checkpoint_interval must equal to validate_interval"self.keep_nbest_models = kwargs.get("keep_nbest_models", 500)self.avg_keep_nbest_models_type = kwargs.get("avg_keep_nbest_models_type", "acc")self.avg_nbest_model = kwargs.get("avg_nbest_model", 10)