Update engines/orpheus_engine.py

engines/orpheus_engine.py

@@ -10,7 +10,7 @@ from typing import Optional, Union
 from RealtimeTTS.engines import BaseEngine, TimingInfo
 
 # Default configuration values
-DEFAULT_API_URL = "http://127.0.0.1:1234/v1/completions"
+DEFAULT_API_URL = "http://127.0.0.1:1234"
 DEFAULT_HEADERS = {"Content-Type": "application/json"}
 DEFAULT_MODEL = "orpheus-3b-0.1-ft"
 DEFAULT_VOICE = "JaKob"
@@ -177,6 +177,7 @@ class OrpheusEngine(BaseEngine):
         payload = {
             "model": self.model,
             "prompt": formatted_prompt,
+            "messages": [{"role": "user", "content": formatted_prompt}],
             "max_tokens": self.max_tokens,
             "temperature": self.temperature,
             "top_p": self.top_p,
@@ -211,7 +212,8 @@ class OrpheusEngine(BaseEngine):
                         try:
                             data = json.loads(data_str)
                             if 'choices' in data and data['choices']:
-                                token_text = data['choices'][0].get('text', '')
+                                delta      = data["choices"][0]
+                                token_text = delta.get("delta", {}).get("content", "")
                                 if token_text:
                                     token_counter += 1
                                     # Print the time it took to get the first token