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
fixbug
游雁9e13f028bc4c
diff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.pyindex 57da105b8..66c416b3d 100644--- a/funasr/models/llm_asr/model.py+++ b/funasr/models/llm_asr/model.py@@ -714,13 +714,17 @@ class LLMASR2(nn.Module):]llm_dtype = kwargs.get("llm_dtype", "fp32")+ if llm_dtype == "fp32":+ llm_dtype = "fp16" if kwargs.get("fp16", False) else llm_dtype+ llm_dtype = "bf16" if kwargs.get("bf16", False) else llm_dtype+dtype_map = {"bf16": torch.bfloat16, "fp16": torch.float16, "fp32": torch.float32}with torch.cuda.amp.autocast(enabled=True if llm_dtype != "fp32" else False, dtype=dtype_map[llm_dtype]):label = contents["assistant"][0]- # self.llm = self.llm.to(dtype_map[llm_dtype])- # inputs_embeds = inputs_embeds.to(dtype_map[llm_dtype])+ self.llm = self.llm.to(dtype_map[llm_dtype])+ inputs_embeds = inputs_embeds.to(dtype_map[llm_dtype])if not kwargs.get("tearchforing", False):