hypaai commited on
Commit
fac0887
·
verified ·
1 Parent(s): 57df8d3

Update handler.py

Browse files
Files changed (1) hide show
  1. handler.py +14 -0
handler.py CHANGED
@@ -64,6 +64,8 @@ class EndpointHandler:
64
  self.padding_token = 128263
65
  self.start_audio_token = 128257 # Start of Audio token
66
  self.end_audio_token = 128258 # End of Audio token
 
 
67
 
68
  logger.info("Handler initialization complete")
69
 
@@ -268,6 +270,10 @@ class EndpointHandler:
268
  """
269
  Main entry point for the handler
270
  """
 
 
 
 
271
  try:
272
  logger.info(f"Received request: {type(data)}")
273
 
@@ -286,3 +292,11 @@ class EndpointHandler:
286
  logger.error(traceback.format_exc())
287
  return {"error": str(e)}
288
 
 
 
 
 
 
 
 
 
 
64
  self.padding_token = 128263
65
  self.start_audio_token = 128257 # Start of Audio token
66
  self.end_audio_token = 128258 # End of Audio token
67
+
68
+ self._warmed_up = False
69
 
70
  logger.info("Handler initialization complete")
71
 
 
270
  """
271
  Main entry point for the handler
272
  """
273
+ # Run warmup only once, the first time __call__ is triggered
274
+ if not self._warmed_up:
275
+ self._warmup()
276
+
277
  try:
278
  logger.info(f"Received request: {type(data)}")
279
 
 
292
  logger.error(traceback.format_exc())
293
  return {"error": str(e)}
294
 
295
+ def _warmup(self):
296
+ try:
297
+ dummy_prompt = "tara: Hello"
298
+ input_ids = self.tokenizer(dummy_prompt, return_tensors="pt").input_ids.to(self.device)
299
+ _ = self.model.generate(input_ids=input_ids, max_new_tokens=5)
300
+ self._warmed_up = True
301
+ except Exception as e:
302
+ print(f"[WARMUP ERROR] {str(e)}")