fix: Choose correct way to generate based on model name

agents-api/agents_api/rec_sum/generate.py

@@ -1,21 +1,25 @@
-from openai import AsyncClient
 from tenacity import retry, stop_after_attempt, wait_fixed
-
-
-client = AsyncClient()
+from agents_api.env import model_inference_url, model_api_key
+from agents_api.model_registry import JULEP_MODELS
+from litellm import acompletion
 
 
 @retry(wait=wait_fixed(2), stop=stop_after_attempt(5))
 async def generate(
     messages: list[dict],
-    client: AsyncClient = client,
     model: str = "gpt-4-turbo",
-    **kwargs
+    **kwargs,
 ) -> dict:
-    result = await client.chat.completions.create(
-        model=model, messages=messages, **kwargs
-    )
+    base_url, api_key = None, None
+    if model in JULEP_MODELS:
+        base_url, api_key = model_inference_url, model_api_key
+        model = f"openai/{model}"
 
-    result = result.choices[0].message.__dict__
+    result = await acompletion(
+        model=model,
+        messages=messages,
+        base_url=base_url,
+        api_key=api_key,
+    )
 
-    return result
+    return result.choices[0].message.json()