'#1823: use float16, not int8, for better precision and ~50% more speed

sepinf-inc · Apr 15, 2024 · b6ec69d · b6ec69d
1 parent cc7b495
commit b6ec69d
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/iped-app/resources/scripts/tasks/WhisperProcess.py b/iped-app/resources/scripts/tasks/WhisperProcess.py
@@ -35,13 +35,13 @@ def main():
         deviceNum = 0
 
     try:
-        model = WhisperModel(modelName, device=deviceId, device_index=deviceNum, cpu_threads=threads, compute_type="int8")
+        model = WhisperModel(modelName, device=deviceId, device_index=deviceNum, cpu_threads=threads, compute_type="float16")
 
     except Exception as e:
         if deviceId != 'cpu':
             # loading on GPU failed (OOM?), try on CPU
             deviceId = 'cpu'
-            model = WhisperModel(model_size_or_path=modelName, device=deviceId, cpu_threads=threads, compute_type="int8")
+            model = WhisperModel(model_size_or_path=modelName, device=deviceId, cpu_threads=threads, compute_type="float16")
         else:
             raise e