Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

decoding

游雁407625a73478
diff --git a/funasr/bin/train_ds.py b/funasr/bin/train_ds.py
index 67ed2ba93..41ecbe4bc 100644
--- a/funasr/bin/train_ds.py
+++ b/funasr/bin/train_ds.py
@@ -124,6 +124,7 @@ def main(**kwargs):
use_ddp=use_ddp,
use_fsdp=use_fsdp,
device=kwargs["device"],
+ excludes=kwargs.get("excludes", None),
output_dir=kwargs.get("output_dir", "./exp"),
**kwargs.get("train_conf"),
)
diff --git a/funasr/train_utils/trainer_ds.py b/funasr/train_utils/trainer_ds.py
index ec76531d6..c97372851 100644
--- a/funasr/train_utils/trainer_ds.py
+++ b/funasr/train_utils/trainer_ds.py
@@ -147,6 +147,10 @@ class Trainer:
self.use_deepspeed = use_deepspeed
self.deepspeed_config = kwargs.get("deepspeed_config", "")
+ self.excludes = kwargs.get("excludes", None)
+ if self.excludes is not None:
+ if isinstance(self.excludes, str):
+ self.excludes = self.excludes.split(",")
def save_checkpoint(
self,
@@ -440,6 +444,12 @@ class Trainer:
src_state = checkpoint["state_dict"]
dst_state = model.state_dict()
for k in dst_state.keys():
+ if excludes is not None:
+ for k_ex in excludes:
+ k_tmp = k.replace("module.", "")
+ if k_tmp.startswith(k_ex):
+ logging.info(f"key: {{k}} matching: {k_ex}, excluded")
+ continue
if not k.startswith("module.") and "module." + k in src_state.keys():
k_ddp = "module." + k
elif k.startswith("module.") and "module." + k not in src_state.keys():