HfApiModel->TransformersModel

app.py

@@ -55,7 +55,7 @@ def create_prompt_for_image_generation(user_prompt: str) -> str:
 	6.	Output Settings: Suggest aspect ratio, output format (e.g., PNG), quality level, and seed for reproducibility.
     Ensure that the generated prompt is logical, descriptive, and written in natural language to maximize compatibility with FLUX-Schnell capabilities.
     Example Input:
-    An image of a serene forest with a small cabin.'
+    An image of a serene forest with a small cabin.
     Example Output:
     In the foreground, a lush green forest floor covered with moss and scattered wildflowers. 
     In the middle ground, a cozy wooden cabin with smoke gently rising from its chimney. 
@@ -66,21 +66,31 @@ def create_prompt_for_image_generation(user_prompt: str) -> str:
     capturing golden hour lighting for soft shadows and warm highlights. 
     The aspect ratio is 1:1, using seed 42 for reproducibility.
     """
-    model = HfApiModel(
-    max_tokens=512,
-    temperature=1.0,
-    model_id='Qwen/Qwen2.5-Coder-32B-Instruct',
-    custom_role_conversions=None,
+    # model = HfApiModel(
+    # max_tokens=512,
+    # temperature=1.0,
+    # model_id='Qwen/Qwen2.5-Coder-32B-Instruct',
+    # custom_role_conversions=None,
+    # )
+    engine = TransformersModel(
+    model_id="Qwen/Qwen2.5-Coder-32B-Instruct",
+    max_new_tokens=512,
     )
+    
     prompt = prefix + user_prompt + '. ' + postfix 
+    messages = [{"role": "user", "content": prompt}]
+    
     try:
-        response = model.generate(
-            prompt=prompt, temperature=1., max_tokens=512)
-        return response['choices'][0]['text'] 
+        # response = model(
+        #     prompt=prompt, temperature=1., max_tokens=512)
+        response = engine(messages, stop_sequences=["END"])
+        # return response['choices'][0]['text'] 
+        return response
 
     except Exception as e:
         return f"Error during LLM call: {str(e)}"
 
+
 final_answer = FinalAnswerTool()
 
 # If the agent does not answer, the model is overloaded, please use another model or the following Hugging Face Endpoint that also contains qwen2.5 coder: