fix syntax error

yzhang123 · yzhang123 · commit c72df2be4aa3 · 2020-02-14T15:49:54.000-08:00
diff --git a/PyTorch/SpeechRecognition/Jasper/trt/README.md b/PyTorch/SpeechRecognition/Jasper/trt/README.md
@@ -229,7 +229,7 @@ NUM_STEPS: Number of inference steps. If -1 runs inference on entire dataset (de
 BATCH_SIZE: data batch size (default: 64)
 NUM_FRAMES: cuts/pads all pre-processed feature tensors to this length. 100 frames ~ 1 second of audio (default: 512)
 FORCE_ENGINE_REBUILD: boolean that indicates whether an already-built TensorRT engine of equivalent precision, batch-size, and number of frames should not be used. Engines are specific to the GPU, library versions, TensorRT versions, and CUDA versions they were built in and cannot be used in a different environment. (default: "true")
-USE_DYNAMIC_SHAPE: if 'yes' uses dynamic shapes (default: ‘yes’)
+USE_DYNAMIC_SHAPE: if 'yes' uses dynamic shapes (default: ‘yes’). Dynamic shape is always preferred since it allows to reuse engines.
 ```
 
 The complete list of parameters available for `trt/scripts/trt_inference.sh` is the same as `trt/scripts/trt_inference_benchmark.sh` only with different default input arguments. In the following, only the parameters with different default values are listed:
diff --git a/PyTorch/SpeechRecognition/Jasper/trt/perfutils.py b/PyTorch/SpeechRecognition/Jasper/trt/perfutils.py
@@ -105,7 +105,7 @@ def get_onnx(path, acoustic_model,  args):
     '''
     
     dynamic_dim = 0
-    if args.dynamic_shape:
+    if not args.static_shape:
         dynamic_dim = 1 if args.transpose else 2
 
 
