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
init param
游雁a7532ceca9ed
diff --git a/funasr/auto/auto_model.py b/funasr/auto/auto_model.pyindex 1abdcea57..a6be69115 100644--- a/funasr/auto/auto_model.py+++ b/funasr/auto/auto_model.py@@ -193,7 +193,7 @@ class AutoModel:path=init_param,ignore_init_mismatch=kwargs.get("ignore_init_mismatch", False),oss_bucket=kwargs.get("oss_bucket", None),- scope_map=kwargs.get("scope_map", "module.,None"),+ scope_map=kwargs.get("scope_map", []),excludes=kwargs.get("excludes", None),)else:diff --git a/funasr/bin/train.py b/funasr/bin/train.pyindex 6650f0a87..569757a6a 100644--- a/funasr/bin/train.py+++ b/funasr/bin/train.py@@ -105,7 +105,7 @@ def main(**kwargs):path=p,ignore_init_mismatch=kwargs.get("ignore_init_mismatch", True),oss_bucket=kwargs.get("oss_bucket", None),- scope_map=kwargs.get("scope_map", "module.,none"),+ scope_map=kwargs.get("scope_map", []),excludes=kwargs.get("excludes", None),)else:diff --git a/funasr/models/llm_asr_nar/model.py b/funasr/models/llm_asr_nar/model.pyindex a61190c1d..94dae4d90 100644--- a/funasr/models/llm_asr_nar/model.py+++ b/funasr/models/llm_asr_nar/model.py@@ -315,7 +315,7 @@ class LLMASRNAR(nn.Module):model_outputs = self.llm(inputs_embeds=inputs_embeds, attention_mask=attention_mask, labels=None)preds = torch.argmax(model_outputs.logits, -1)text = tokenizer.batch_decode(preds, add_special_tokens=False, skip_special_tokens=True)- text = text.split(': "\n')[-1]+ text = text.split(': \n')[-1]# preds = torch.argmax(model_outputs.logits, -1)ibest_writer = Nonediff --git a/funasr/train_utils/load_pretrained_model.py b/funasr/train_utils/load_pretrained_model.pyindex ab022dc9b..7a8c37459 100644--- a/funasr/train_utils/load_pretrained_model.py+++ b/funasr/train_utils/load_pretrained_model.py@@ -45,7 +45,7 @@ def load_pretrained_model(ignore_init_mismatch: bool=True,map_location: str = "cpu",oss_bucket=None,- scope_map="module.:none",+ scope_map=[],excludes=None,ignore_mismatch=False,**kwargs,@@ -75,6 +75,7 @@ def load_pretrained_model(if isinstance(scope_map, str):scope_map = scope_map.split(",")+ scope_map += ["module.", "None"]for k in dst_state.keys():@@ -86,9 +87,12 @@ def load_pretrained_model(for i in range(0, len(scope_map), 2):src_prefix = scope_map[i] if scope_map[i].lower() != "none" else ""dst_prefix = scope_map[i+1] if scope_map[i+1].lower() != "none" else ""-- if k.startswith(dst_prefix) and k.replace(dst_prefix, src_prefix) in src_state.keys():- k_src = k.replace(dst_prefix, src_prefix)++ if dst_prefix == "" and (src_prefix + k) in src_state.keys():+ k_src = src_prefix + k+ print(f"init param, map: {k} from {k_src} in ckpt")+ elif k.startswith(dst_prefix) and k.replace(dst_prefix, src_prefix, 1) in src_state.keys():+ k_src = k.replace(dst_prefix, src_prefix, 1)print(f"init param, map: {k} from {k_src} in ckpt")if k_src in src_state.keys():