Try use local model or OpenAI model and Streamlit example

Gradio_UI.py

@@ -157,7 +157,7 @@ def pull_messages_from_step(
 
 
 def stream_to_gradio(
-    agent,
+    agent: MultiStepAgent,
     task: str,
     reset_agent_memory: bool = False,
     additional_args: Optional[dict] = None,
@@ -175,8 +175,21 @@ def stream_to_gradio(
     for step_log in agent.run(
         task, stream=True, reset=reset_agent_memory, additional_args=additional_args
     ):
+        if hasattr(step_log, "error") and step_log.error:
+            # AttributeError: 'AgentText' object has no attribute 'error'
+            # BUG: (this is model endpoint issue)
+            # Error in generating model output:
+            # (Request ID: UkBnX7)
+            # Bad request:
+            # Bad Request: Invalid state
+            yield gr.ChatMessage(
+                role="assistant", content=f"**Error:** {str(step_log.error)}"
+            )
+            break
+
         # Track tokens if model provides them
         if hasattr(agent.model, "last_input_token_count"):
+            # BUG: TypeError: unsupported operand type(s) for +=: 'int' and 'NoneType'
             total_input_tokens += agent.model.last_input_token_count
             total_output_tokens += agent.model.last_output_token_count
             if isinstance(step_log, ActionStep):

README.md

@@ -23,5 +23,10 @@ Check out the configuration reference at https://huggingface.co/docs/hub/spaces-
 
 ```bash
 # BUG: somehow failed to inference locally
+# Gradio
 python app.py
+# Streamlit
+streamlit run streamlit_app.py
+# Chainlit
+chainlit run chainlit_app.py
 ```

app.py

@@ -1,6 +1,13 @@
-from smolagents import CodeAgent, HfApiModel, load_tool, tool
+from smolagents import (
+    CodeAgent,
+    HfApiModel,
+    TransformersModel,
+    OpenAIServerModel,
+    load_tool,
+    tool,
+)
+import os
 import datetime
-import requests
 import pytz
 import yaml
 from tools import (
@@ -9,12 +16,10 @@ from tools import (
     DuckDuckGoSearchTool,  # This is also built-in in smolagents
 )
 
-from Gradio_UI import GradioUI
-
 
 # Below is an example of a tool that does nothing. Amaze us with your creativity !
 @tool
-def my_cutom_tool(
+def my_custom_tool(
     arg1: str, arg2: int
 ) -> str:  # it's import to specify the return type
     # Keep this format for the description / args / args description but feel free to modify the tool
@@ -43,12 +48,38 @@ def get_current_time_in_timezone(timezone: str) -> str:
 
 
 final_answer = FinalAnswerTool()
-model = HfApiModel(
-    max_tokens=2096,
-    temperature=0.5,
-    model_id="https://wxknx1kg971u7k1n.us-east-1.aws.endpoints.huggingface.cloud",  # it is possible that this model may be overloaded
-    custom_role_conversions=None,
-)
+
+if IS_IN_HF_SPACE := os.getenv("SPACE_ID"):
+    model_id = (
+        # BUG: Seems we are failed to call this model => bad request
+        "https://wxknx1kg971u7k1n.us-east-1.aws.endpoints.huggingface.cloud"  # it is possible that this model may be overloaded
+        if True
+        # BUG: Model not loaded on the server: https://api-inference.huggingface.co/models/Qwen/Qwen2.5-Coder-32B-Instruct/v1/chat/completions. Please retry with a higher timeout (current: 120)
+        # BUG: TooManyRequests: Please log in or use a HF access token
+        else "Qwen/Qwen2.5-Coder-32B-Instruct"  # The default value of HfApiModel
+    )
+
+    model = HfApiModel(
+        max_tokens=2096,
+        temperature=0.5,
+        model_id=model_id,
+        custom_role_conversions=None,
+    )
+
+else:
+    from dotenv import load_dotenv
+
+    # NOTE: to load the Hugging Face API Key
+    curr_dir = os.path.dirname(os.path.abspath(__file__))
+    load_dotenv(os.path.join(curr_dir, "../../.env"))
+
+    if OPENAI_API_KEY := os.getenv("OPENAI_API_KEY"):
+        model = OpenAIServerModel(model_id="gpt-3.5-turbo", api_key=OPENAI_API_KEY)
+    else:
+        # NOTE: this model is not good enough for agent
+        model = TransformersModel(
+            model_id="HuggingFaceTB/SmolLM2-1.7B-Instruct", trust_remote_code=True
+        )
 
 
 # Import tool from Hub
@@ -64,6 +95,8 @@ agent = CodeAgent(
     model=model,
     tools=[
         final_answer,
+        my_custom_tool,
+        get_current_time_in_timezone,
         image_generation_tool,
         duckduckgo_tool,
         visit_webpage_tool,
@@ -77,5 +110,7 @@ agent = CodeAgent(
     prompt_templates=prompt_templates,
 )
 
+if __name__ == "__main__":
+    from Gradio_UI import GradioUI
 
-GradioUI(agent).launch()
+    GradioUI(agent).launch()

requirements.txt

@@ -1,5 +1,5 @@
 markdownify
-smolagents
+smolagents[openai]
 requests
 duckduckgo_search
 pandas
@@ -7,4 +7,9 @@ pandas
 # Local development
 gradio
 # ImportError: Using SOCKS proxy, but the 'socksio' package is not installed. Make sure to install httpx using `pip install httpx[socks]`
-httpx[socks]
+httpx[socks]
+
+# Try different UI
+streamlit
+watchdog
+chainlit

streamlit_app.py

@@ -0,0 +1,58 @@
+import streamlit as st
+
+# Import your agent and streaming function.
+# For example, if your agent code is in agent_app.py:
+from app import agent  # your CodeAgent instance
+from Gradio_UI import (
+    stream_to_gradio,
+)  # re-use the generator that yields gr.ChatMessage
+
+# (Optionally, if you want to avoid Gradio-specific types you can write your own streaming generator.)
+
+st.set_page_config(page_title="CodeAgent Chat", layout="wide")
+st.title("CodeAgent Chat (Streamlit)")
+
+# Initialize session state for chat history.
+if "chat_history" not in st.session_state:
+    st.session_state.chat_history = []
+
+
+def display_chat():
+    """Display the chat history in the app."""
+    for message in st.session_state.chat_history:
+        role = message.get("role", "assistant")
+        content = message.get("content", "")
+        if role == "user":
+            st.markdown(f"**User:** {content}")
+        else:
+            st.markdown(f"**Assistant:** {content}")
+
+
+# Main chat container.
+chat_container = st.container()
+with chat_container:
+    display_chat()
+
+# TODO: use `st.chat_input`
+
+# User input.
+user_input = st.text_input("Enter your message:", key="input_text")
+if st.button("Send") and user_input:
+    # Append the user message to the history.
+    st.session_state.chat_history.append({"role": "user", "content": user_input})
+    with chat_container:
+        display_chat()
+
+    # Stream the agent responses.
+    # (Here we are reusing your existing streaming generator.
+    #  Note that gr.ChatMessage objects have attributes "role" and "content".)
+    placeholder = st.empty()  # if you want to update a placeholder
+    for msg in stream_to_gradio(agent, user_input, reset_agent_memory=False):
+        # Extract role and content.
+        role = msg.role
+        # For content that is not a plain string (e.g. images or audio), you might need to
+        # add extra handling. Here we simply convert it to a string.
+        content = msg.content if isinstance(msg.content, str) else str(msg.content)
+        st.session_state.chat_history.append({"role": role, "content": content})
+        with chat_container:
+            display_chat()