fixed to device for inputs

handler.py

@@ -81,39 +81,30 @@ class EndpointHandler:
             data (:dict:):
                 The payload with the text prompt and generation parameters.
 
-       Returns: wav file in bytes
         """
 
-        # inputs = data.pop("inputs", data)
-        # params = data.pop("parameters", None)
+        inputs = data.pop("inputs", data)
+        params = data.pop("parameters", None)
 
-        # inputs = self.processor(
-        #     text=[inputs],
-        #     padding=True,
-        #     return_tensors="pt"
-        # ).to('cuda')
+        inputs = self.processor(
+            text=[inputs],
+            padding=True,
+            return_tensors="pt"
+        ).to('cuda')
 
-        # params = create_params(params, self.model.config.audio_encoder.frame_rate)
+        params = create_params(params, self.model.config.audio_encoder.frame_rate)
 
-        # with torch.cuda.amp.autocast():
-        #     outputs = self.model.generate(**inputs, **params)
+        with torch.cuda.amp.autocast():
+            outputs = self.model.generate(**inputs.to('cuda'), do_sample=True, guidance_scale=3, max_new_tokens=256)
 
-        # pred = outputs[0, 0].cpu().numpy()
-        # sr = 32000
+        pred = outputs[0, 0].cpu().numpy()
+        sr = self.model.config.audio_encoder.sampling_rate
 
-        # try:
-        #      sr = self.model.config.audio_encoder.sampling_rate
+        wav_buffer = io.BytesIO()
+        wavfile.write(wav_buffer, rate=sr, data=pred)
+        wav_data = wav_buffer.getvalue()
 
-        # except:
-        #      sr = 32000
-
-        # wav_buffer = io.BytesIO()
-        # wavfile.write(wav_buffer, rate=sr, data=pred)
-        # wav_data = wav_buffer.getvalue()
-
-        # base64_encoded_wav = base64.b64encode(wav_data).decode('utf-8')
-
-        base64_encoded_wav = sine_to_base64()
+        base64_encoded_wav = base64.b64encode(wav_data).decode('utf-8')
         return  [{"audio": base64_encoded_wav}]