Upload folder using huggingface_hub

app.py

@@ -20,12 +20,41 @@ def patch_tinytroupe():
     if os.path.exists(path):
         with open(path, "r") as f:
             content = f.read()
-        # Replace alias-large with alias-fast in the client() function
-        content = content.replace('config_manager.update("model", "alias-large")', 'config_manager.update("model", "alias-fast")')
-        content = content.replace('config_manager.update("reasoning_model", "alias-large")', 'config_manager.update("reasoning_model", "alias-fast")')
+        
+        # Ensure alias-large is used (it is the default in the client() function in the branch)
+        
+        # Handle 502 errors on Helmholtz by waiting 35 seconds
+        old_502_block = """                # Temporary fallback for 502 errors on Helmholtz
+                if isinstance(e, openai.APIStatusError) and e.status_code == 502 and isinstance(self, HelmholtzBlabladorClient):
+                    logger.warning("Helmholtz API returned a 502 error. Temporarily falling back to OpenAI for this request.")
+                    try:
+                        fallback_client = _get_client_for_api_type("openai")
+                        fallback_chat_api_params = chat_api_params.copy()
+                        fallback_chat_api_params["model"] = "gpt-4o-mini"
+                        fallback_chat_api_params["max_tokens"] = 16384
+
+                        response = fallback_client._raw_model_call(fallback_chat_api_params["model"], fallback_chat_api_params)
+
+                        if enable_pydantic_model_return:
+                            return utils.to_pydantic_or_sanitized_dict(fallback_client._raw_model_response_extractor(response), model=response_format)
+                        else:
+                            return utils.sanitize_dict(fallback_client._raw_model_response_extractor(response))
+                    except Exception as fallback_e:
+                        logger.error(f"Fallback to OpenAI also failed: {fallback_e}")"""
+        
+        new_502_block = """                # Handle 502 errors on Helmholtz by waiting 35 seconds
+                if isinstance(e, openai.APIStatusError) and e.status_code == 502 and isinstance(self, HelmholtzBlabladorClient):
+                    logger.warning("Helmholtz API returned a 502 error. Waiting 35 seconds before retrying...")
+                    time.sleep(35)"""
+        
+        if old_502_block in content:
+            content = content.replace(old_502_block, new_502_block)
+        else:
+            print("Could not find the 502 block to patch.")
+            
         with open(path, "w") as f:
             f.write(content)
-        print("TinyTroupe patched to use alias-fast.")
+        print("TinyTroupe patched to handle 502 errors with 35s wait.")
 
 clone_tinytroupe()
 
@@ -96,9 +125,9 @@ def get_repo_branches(repo_full_name):
         return ["main"]
 
 def generate_personas(theme, customer_profile, num_personas):
-    # Override model if alias-large is failing
-    config_manager.update("model", "alias-fast")
-    config_manager.update("reasoning_model", "alias-fast")
+    # Ensure alias-large is used
+    config_manager.update("model", "alias-large")
+    config_manager.update("reasoning_model", "alias-large")
     
     context = f"A company related to {theme}. Target customers: {customer_profile}"
     
@@ -164,8 +193,8 @@ def generate_tasks(theme, customer_profile):
     Return the tasks as a JSON list of strings in the format: {{"tasks": ["task1", "task2", ...]}}
     """
     
-    # Try alias-large first, then alias-fast
-    for model_name in ["alias-large", "alias-fast"]:
+    # Use alias-large
+    for model_name in ["alias-large"]:
         try:
             response = client.chat.completions.create(
                 model=model_name,

config.ini

@@ -18,7 +18,7 @@ BLABLADOR_ENDPOINT=https://api.helmholtz-blablador.fz-juelich.de/v1
 #
 
 # The main text generation model, used for agent responses
-MODEL=alias-fast
+MODEL=alias-large
 
 # Reasoning model is used when precise reasoning is required, such as when computing detailed analyses of simulation properties.
 REASONING_MODEL=alias-large