成功推理（流式）

examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py

@@ -115,7 +115,7 @@ from nemo.collections.asr.parts.submodules.rnnt_decoding import RNNTDecodingConf
 from nemo.collections.asr.parts.utils.manifest_utils import read_manifest
 from nemo.collections.asr.parts.utils.rnnt_utils import Hypothesis
 from nemo.collections.asr.parts.utils.streaming_utils import CacheAwareStreamingAudioBuffer
-from nemo.collections.asr.parts.utils.transcribe_utils import get_inference_device, get_inference_dtype, setup_model
+from nemo.collections.asr.parts.utils.transcribe_utils import setup_model
 from nemo.core.config import hydra_runner
 from nemo.utils import logging
 
@@ -309,7 +309,15 @@ def main(cfg: TranscriptionConfig):
         pl.seed_everything(cfg.random_seed)
 
     # setup device
-    device = get_inference_device(cuda=cfg.cuda, allow_mps=cfg.allow_mps)
+    if cfg.cuda is None:
+        device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+    elif cfg.cuda < 0:
+        device = torch.device('cpu')
+    else:
+        device = torch.device(f'cuda:{cfg.cuda}')
+        
+    if cfg.allow_mps and getattr(torch.backends, "mps", None) is not None and torch.backends.mps.is_available():
+        device = torch.device('mps')
 
     if (cfg.compute_dtype is not None and cfg.compute_dtype != "float32") and cfg.amp:
         raise ValueError("amp=true is mutually exclusive with a compute_dtype other than float32")
@@ -321,7 +329,10 @@ def main(cfg: TranscriptionConfig):
         # with amp model weights required to be in float32
         compute_dtype = torch.float32
     else:
-        compute_dtype = get_inference_dtype(compute_dtype=cfg.compute_dtype, device=device)
+        if cfg.compute_dtype is None:
+            compute_dtype = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32
+        else:
+            compute_dtype = getattr(torch, cfg.compute_dtype)
 
     if compute_dtype != torch.float32:
         # NB: cache-aware models do not currently work with compute_dtype != float32
@@ -405,8 +416,11 @@ def main(cfg: TranscriptionConfig):
                 streaming_buffer=streaming_buffer,
                 compute_dtype=compute_dtype,
                 compare_vs_offline=cfg.compare_vs_offline,
+                debug_mode=cfg.debug_mode,
                 pad_and_drop_preencoded=cfg.pad_and_drop_preencoded,
             )
+            # return early for single file mode
+            return
         else:
             # stream audio files in a manifest file in batched mode
             all_streaming_tran = []
@@ -486,4 +500,13 @@ def main(cfg: TranscriptionConfig):
 
 
 if __name__ == '__main__':
+    import sys
+    sys.argv.extend([
+        'model_path=results/NeMo_Ja_FastConformer_Streaming/checkpoints/NeMo_Ja_FastConformer_Streaming.nemo',
+        "audio_file=E:/huggingface_echodict/NeMo/data/common_voice_11_0/ja/train/ja_train_0/common_voice_ja_25372057.wav",
+        'batch_size=16',
+        'compare_vs_offline=false',
+        'debug_mode=true',
+        'amp=true'
+    ])
     main()

readme.txt

@@ -236,6 +236,10 @@ python examples/asr/transcribe_speech.py
     流式训练改三个参数，成功训练
 
 
+成功推理（流式）
+python examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py
+
+
 
 
 如果要训练一个 ASR 语音识别大模型，只用一条数据，它既是训练集，又是验证集，又是测试集。这样它正确率能正 100% 吗