Full memory storing and retrieval for conversation agent. Update gral prompts. Add custom inference model.

backend/inference.py

@@ -11,8 +11,16 @@ class LlamaMessage:
         self.role = role
         self.content = content
 
-async def replicate_llama_completion(app, messages):
-    # Llama 3 instruct
+
+class Models:
+    llama_8b = "meta/meta-llama-3-8b-instruct"
+    llama_70b = "meta/meta-llama-3-70b-instruct"
+
+async def replicate_llama_completion(app, messages, model=None):
+    if model is None:
+        model = Models.llama_8b
+
+    # Llama instruct
     full_prompt = "<|begin_of_text|>"
     for message in messages:
         header_name = message.role if message.role != "agent" else "assistant"
@@ -24,7 +32,7 @@ async def replicate_llama_completion(app, messages):
         target=run_replicate_inference_loop,
         kwargs={
             'replicate_client': app.replicate_client,
-            'model': LLAMA3_MODEL,
+            'model': model,
             'prompt': full_prompt,
             'content_stream': content_stream,
         }