Upload folder using huggingface_hub

README.md

@@ -15,11 +15,12 @@ This is a Gradio demo of the WebVoyager agent, which can complete user instructi
 
 ## Setup
 
-To use this space, you must set your OpenAI API key as a secret.
+To use this space, you must set your `BLABLADOR_API_KEY` and `OPENAI_API_KEY` as secrets.
 
 1.  Go to the "Settings" tab of this space.
 2.  Scroll down to the "Secrets" section.
 3.  Click on "New secret" and add your `BLABLADOR_API_KEY`.
+4.  Click on "New secret" and add your `OPENAI_API_KEY`.
 
 ## How to use
 

app.py

@@ -68,7 +68,7 @@ def run_script_for_gradio(url, task):
             max_iter=5,
             api_key=os.environ.get("BLABLADOR_API_KEY"),
             api_base_url="https://api.helmholtz-blablador.fz-juelich.de/v1",
-            api_model="alias-fats",
+            api_model="alias-fast",
             output_dir=os.path.join(temp_dir, 'results'),
             seed=None,
             max_attached_imgs=1,

run.py

@@ -85,19 +85,19 @@ def format_msg(it, init_msg, pdf_obs, warn_obs, web_img_b64, web_text):
             ]
         }
 
-def call_gpt4v_api(args, openai_client, messages):
+def call_gpt4v_api(args, client_blablador, client_openai, messages):
     retry_times = 0
     wait_time = 5  # Initial wait time
     while True:
         try:
             if not args.text_only:
                 logging.info('Calling gpt4v API...')
-                openai_response = openai_client.chat.completions.create(
-                    model=args.api_model, messages=messages, max_tokens=1000, seed=args.seed
+                openai_response = client_openai.chat.completions.create(
+                    model="gpt-4-vision-preview", messages=messages, max_tokens=1000, seed=args.seed
                 )
             else:
                 logging.info('Calling gpt4 API...')
-                openai_response = openai_client.chat.completions.create(
+                openai_response = client_blablador.chat.completions.create(
                     model='alias-large', messages=messages, max_tokens=1000, seed=args.seed, timeout=30
                 )
 
@@ -253,7 +253,8 @@ def webvoyager_run(args, task, task_dir):
     """
     A generator function that yields log messages for each step of the WebVoyager run.
     """
-    client = OpenAI(api_key=args.api_key, base_url=args.api_base_url, http_client=httpx.Client(trust_env=False))
+    client_blablador = OpenAI(api_key=args.api_key, base_url=args.api_base_url, http_client=httpx.Client(trust_env=False))
+    client_openai = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"), http_client=httpx.Client(trust_env=False))
     options = driver_config(args)
     setup_logger(task_dir)
     
@@ -278,7 +279,7 @@ def webvoyager_run(args, task, task_dir):
         # Log the user message
         yield json.dumps(curr_msg)
 
-        for status, result in call_gpt4v_api(args, client, messages):
+        for status, result in call_gpt4v_api(args, client_blablador, client_openai, messages):
             if status == "status":
                 yield json.dumps({"status": result})
             elif status == "result":

utils.py

@@ -370,7 +370,7 @@ def get_pdf_retrieval_ans_from_assistant(client, pdf_path, task):
     logging.info("Create assistant...")
     assistant = client.beta.assistants.create(
         instructions="You are a helpful assistant that can analyze the content of a PDF file and give an answer that matches the given task, or retrieve relevant content that matches the task.",
-        model="alias-fats",
+        model="alias-fast",
         tools=[{"type": "retrieval"}],
         file_ids=[file.id]
     )