Update handler.py

handler.py

@@ -19,81 +19,6 @@ rather than the custom audio generation handler you've defined.
 Create a `handler.py` file with your custom handler code:
 """
 
-# import torch
-# import numpy as np
-
-# from transformers import AutoModelForCausalLM, AutoTokenizer
-
-# class EndpointHandler():
-#     def __init__(self, path=""):
-        
-#         # Load the models and tokenizer
-#         self.model = AutoModelForCausalLM.from_pretrained(
-#             "hypaai/Hypa_Orpheus-3b-0.1-ft-unsloth-merged_16bit",
-#             torch_dtype=torch.bfloat16
-#         )
-#         self.tokenizer = AutoTokenizer.from_pretrained("hypaai/Hypa_Orpheus-3b-0.1-ft-unsloth-merged_16bit")
-
-#         # Move to devices
-#         self.device = "cuda" if torch.cuda.is_available() else "cpu"
-#         self.model.to(self.device)
-
-#         # Special tokens
-#         self.start_token = torch.tensor([[128259]], dtype=torch.int64)  # Start of human
-#         self.end_tokens = torch.tensor([[128009, 128260]], dtype=torch.int64)  # End of text, End of human
-#         self.padding_token = 128263
-#         self.start_audio_token = 128257  # Start of Audio token
-#         self.end_audio_token = 128258  # End of Audio token
-
-
-#     def __call__(self, data):
-#         """
-#         Main entry point for the handler
-#         """
-        
-#         # Preprocess input
-#         if isinstance(data, dict) and "inputs" in data:
-#             text = data["inputs"]
-#             parameters = data.get("parameters", {})
-#         else:
-#             text = data
-#             parameters = {}
-        
-#         # Extract parameters from request
-#         voice = parameters.get("voice", "tara")
-#         temperature = float(parameters.get("temperature", 0.6))
-#         top_p = float(parameters.get("top_p", 0.95))
-#         max_new_tokens = int(parameters.get("max_new_tokens", 1200))
-#         repetition_penalty = float(parameters.get("repetition_penalty", 1.1))
-
-#         # Format prompt with voice
-#         prompt = f"{voice}: {text}"
-
-#         # Tokenize
-#         input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids
-
-#         # Add special tokens
-#         modified_input_ids = torch.cat([self.start_token, input_ids, self.end_tokens], dim=1)
-
-#         # No need for padding as we're processing a single sequence
-#         input_ids = modified_input_ids.to(self.device)
-#         attention_mask = torch.ones_like(input_ids)
-
-#         # Forward pass through the model
-#         generated_ids = self.model.generate(
-#             input_ids=input_ids,
-#             attention_mask=attention_mask,
-#             max_new_tokens=max_new_tokens,
-#             do_sample=True,
-#             temperature=temperature,
-#             top_p=top_p,
-#             repetition_penalty=repetition_penalty,
-#             num_return_sequences=1,
-#             eos_token_id=self.end_audio_token,
-#         )
-        
-#         return generated_ids
-# Code from your original handler, but with some fixes
 import os
 import torch
 import numpy as np
@@ -305,7 +230,7 @@ class EndpointHandler:
         logger.info(f"Audio encoded as base64, length: {len(audio_b64)}")
         
         return {
-            "generated_ids": generated_ids.tolist(), #OOO 05102025
+            "audio_sample": audio_sample,
             "audio_b64": audio_b64,
             "sample_rate": 24000
         }