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2518f03d20ca
diff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.py
index 2a55cd6bb..fb0bee3ca 100644
--- a/funasr/models/llm_asr/model.py
+++ b/funasr/models/llm_asr/model.py
@@ -427,30 +427,36 @@ class LLMASR2(nn.Module):
self.audio_encoder = audio_encoder
# llm
- hub = llm_conf.get("hub", "hf")
self.llm = None
- if hub == "hf":
- from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig
- init_param_path = llm_conf.get("init_param_path", "vicuna-7b-v1.5")
+ from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig
- model = AutoModelForCausalLM.from_pretrained(
- init_param_path,
- load_in_8bit=None,
- device_map=None,
- use_cache=None,
- )
- freeze = llm_conf.get("freeze", True)
- if freeze:
- for name, param in model.named_parameters():
- param.requires_grad = False
- model.eval()
- self.llm = model
+ init_param_path = llm_conf.get("init_param_path", "vicuna-7b-v1.5")
+
+ model = AutoModelForCausalLM.from_pretrained(
+ init_param_path,
+ load_in_8bit=None,
+ device_map=None,
+ use_cache=None,
+ )
+ freeze = llm_conf.get("freeze", True)
+ if freeze:
+ for name, param in model.named_parameters():
+ param.requires_grad = False
+ model.eval()
+ self.llm = model
+ llm_dim = model.get_input_embeddings().weight.shape[-1]
# adaptor
adaptor_class = tables.adaptor_classes.get(audio_adaptor)
audio_adaptor_conf["encoder_dim"] = audio_encoder_output_size
+ audio_adaptor_conf["llm_dim"] = llm_dim
audio_adaptor = adaptor_class(**audio_adaptor_conf)
+ init_param_path = audio_adaptor_conf.get("init_param_path", None)
+ if init_param_path is not None:
+ src_state = torch.load(init_param_path, map_location="cpu")
+ flag = audio_adaptor.load_state_dict(src_state, strict=False)
+ logging.info(f"Loading audio_adaptor ckpt: {init_param_path}, status: {flag}")
self.audio_adaptor = audio_adaptor