feat: implement completions model fallback to Mistral Large 2 in forge_execute

backend.py

@@ -1793,12 +1793,29 @@ async def forge_execute(req: ForgeExecuteRequest, authorization: str = Header(No
         
         for round_num in range(MAX_TOOL_ROUNDS + 1):
             await rate_limiter.wait_for_nim()
-            response = await nim_client.chat.completions.create(
-                model=RECOMMENDED_MODEL,
-                messages=final_messages,
-                tools=TOOLS,
-                tool_choice="auto"
-            )
+            response = None
+            try:
+                response = await nim_client.chat.completions.create(
+                    model=RECOMMENDED_MODEL,
+                    messages=final_messages,
+                    tools=TOOLS,
+                    tool_choice="auto"
+                )
+            except Exception as e:
+                log_activity(f"[Forge Fallback] NIM completions failed: {e}. Retrying with Mistral Large...")
+                try:
+                    if mistral_client:
+                        response = await mistral_client.chat.completions.create(
+                            model="mistral-large-latest",
+                            messages=final_messages,
+                            tools=TOOLS,
+                            tool_choice="auto"
+                        )
+                    else:
+                        raise Exception("Mistral client not initialized")
+                except Exception as e2:
+                    log_activity(f"[Forge Fallback Error] Mistral completions failed too: {e2}")
+                    raise e2
             
             msg = response.choices[0].message
             tool_calls_payload = None