Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,75 +1,143 @@
|
|
| 1 |
import os
|
| 2 |
-
import
|
| 3 |
import requests
|
| 4 |
-
import
|
|
|
|
|
|
|
| 5 |
import pandas as pd
|
| 6 |
|
| 7 |
-
#
|
| 8 |
-
# --- Constants ---
|
| 9 |
-
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 10 |
-
|
| 11 |
-
# --- Basic Agent Definition ---
|
| 12 |
-
# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
|
| 13 |
-
class BasicAgent:
|
| 14 |
-
def __init__(self):
|
| 15 |
-
print("BasicAgent initialized.")
|
| 16 |
-
def __call__(self, question: str) -> str:
|
| 17 |
-
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
| 18 |
-
fixed_answer = "This is a default answer."
|
| 19 |
-
print(f"Agent returning fixed answer: {fixed_answer}")
|
| 20 |
-
return fixed_answer
|
| 21 |
-
|
| 22 |
-
def run_and_submit_all( profile: gr.OAuthProfile | None):
|
| 23 |
-
"""
|
| 24 |
-
Fetches all questions, runs the BasicAgent on them, submits all answers,
|
| 25 |
-
and displays the results.
|
| 26 |
-
"""
|
| 27 |
-
# --- Determine HF Space Runtime URL and Repo URL ---
|
| 28 |
-
space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
|
| 29 |
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
questions_url = f"{api_url}/questions"
|
| 39 |
-
submit_url = f"{api_url}/submit"
|
| 40 |
-
|
| 41 |
-
# 1. Instantiate Agent ( modify this part to create your agent)
|
| 42 |
-
try:
|
| 43 |
-
agent = BasicAgent()
|
| 44 |
-
except Exception as e:
|
| 45 |
-
print(f"Error instantiating agent: {e}")
|
| 46 |
-
return f"Error initializing agent: {e}", None
|
| 47 |
-
# In the case of an app running as a hugging Face space, this link points toward your codebase ( usefull for others so please keep it public)
|
| 48 |
-
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 49 |
-
print(agent_code)
|
| 50 |
|
| 51 |
-
|
| 52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
try:
|
| 54 |
-
|
| 55 |
-
response.raise_for_status()
|
| 56 |
-
questions_data = response.json()
|
| 57 |
-
if not questions_data:
|
| 58 |
-
print("Fetched questions list is empty.")
|
| 59 |
-
return "Fetched questions list is empty or invalid format.", None
|
| 60 |
-
print(f"Fetched {len(questions_data)} questions.")
|
| 61 |
-
except requests.exceptions.RequestException as e:
|
| 62 |
-
print(f"Error fetching questions: {e}")
|
| 63 |
-
return f"Error fetching questions: {e}", None
|
| 64 |
-
except requests.exceptions.JSONDecodeError as e:
|
| 65 |
-
print(f"Error decoding JSON response from questions endpoint: {e}")
|
| 66 |
-
print(f"Response text: {response.text[:500]}")
|
| 67 |
-
return f"Error decoding server response for questions: {e}", None
|
| 68 |
except Exception as e:
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
# 3. Run your Agent
|
| 73 |
results_log = []
|
| 74 |
answers_payload = []
|
| 75 |
print(f"Running agent on {len(questions_data)} questions...")
|
|
@@ -77,120 +145,12 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
|
|
| 77 |
task_id = item.get("task_id")
|
| 78 |
question_text = item.get("question")
|
| 79 |
if not task_id or question_text is None:
|
| 80 |
-
|
| 81 |
-
continue
|
| 82 |
try:
|
| 83 |
-
submitted_answer = agent(question_text)
|
| 84 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 85 |
-
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 86 |
except Exception as e:
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
if not answers_payload:
|
| 91 |
-
print("Agent did not produce any answers to submit.")
|
| 92 |
-
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 93 |
-
|
| 94 |
-
# 4. Prepare Submission
|
| 95 |
-
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
| 96 |
-
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
| 97 |
-
print(status_update)
|
| 98 |
-
|
| 99 |
-
# 5. Submit
|
| 100 |
-
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
| 101 |
-
try:
|
| 102 |
-
response = requests.post(submit_url, json=submission_data, timeout=60)
|
| 103 |
-
response.raise_for_status()
|
| 104 |
-
result_data = response.json()
|
| 105 |
-
final_status = (
|
| 106 |
-
f"Submission Successful!\n"
|
| 107 |
-
f"User: {result_data.get('username')}\n"
|
| 108 |
-
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
| 109 |
-
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
| 110 |
-
f"Message: {result_data.get('message', 'No message received.')}"
|
| 111 |
-
)
|
| 112 |
-
print("Submission successful.")
|
| 113 |
-
results_df = pd.DataFrame(results_log)
|
| 114 |
-
return final_status, results_df
|
| 115 |
-
except requests.exceptions.HTTPError as e:
|
| 116 |
-
error_detail = f"Server responded with status {e.response.status_code}."
|
| 117 |
-
try:
|
| 118 |
-
error_json = e.response.json()
|
| 119 |
-
error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
|
| 120 |
-
except requests.exceptions.JSONDecodeError:
|
| 121 |
-
error_detail += f" Response: {e.response.text[:500]}"
|
| 122 |
-
status_message = f"Submission Failed: {error_detail}"
|
| 123 |
-
print(status_message)
|
| 124 |
-
results_df = pd.DataFrame(results_log)
|
| 125 |
-
return status_message, results_df
|
| 126 |
-
except requests.exceptions.Timeout:
|
| 127 |
-
status_message = "Submission Failed: The request timed out."
|
| 128 |
-
print(status_message)
|
| 129 |
-
results_df = pd.DataFrame(results_log)
|
| 130 |
-
return status_message, results_df
|
| 131 |
-
except requests.exceptions.RequestException as e:
|
| 132 |
-
status_message = f"Submission Failed: Network error - {e}"
|
| 133 |
-
print(status_message)
|
| 134 |
-
results_df = pd.DataFrame(results_log)
|
| 135 |
-
return status_message, results_df
|
| 136 |
-
except Exception as e:
|
| 137 |
-
status_message = f"An unexpected error occurred during submission: {e}"
|
| 138 |
-
print(status_message)
|
| 139 |
-
results_df = pd.DataFrame(results_log)
|
| 140 |
-
return status_message, results_df
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
# --- Build Gradio Interface using Blocks ---
|
| 144 |
-
with gr.Blocks() as demo:
|
| 145 |
-
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 146 |
-
gr.Markdown(
|
| 147 |
-
"""
|
| 148 |
-
**Instructions:**
|
| 149 |
-
|
| 150 |
-
1. Please clone this space, then modify the code to define your agent's logic, the tools, the necessary packages, etc ...
|
| 151 |
-
2. Log in to your Hugging Face account using the button below. This uses your HF username for submission.
|
| 152 |
-
3. Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
|
| 153 |
-
|
| 154 |
-
---
|
| 155 |
-
**Disclaimers:**
|
| 156 |
-
Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
|
| 157 |
-
This space provides a basic setup and is intentionally sub-optimal to encourage you to develop your own, more robust solution. For instance for the delay process of the submit button, a solution could be to cache the answers and submit in a seperate action or even to answer the questions in async.
|
| 158 |
-
"""
|
| 159 |
-
)
|
| 160 |
-
|
| 161 |
-
gr.LoginButton()
|
| 162 |
-
|
| 163 |
-
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 164 |
-
|
| 165 |
-
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 166 |
-
# Removed max_rows=10 from DataFrame constructor
|
| 167 |
-
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 168 |
-
|
| 169 |
-
run_button.click(
|
| 170 |
-
fn=run_and_submit_all,
|
| 171 |
-
outputs=[status_output, results_table]
|
| 172 |
-
)
|
| 173 |
-
|
| 174 |
-
if __name__ == "__main__":
|
| 175 |
-
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
| 176 |
-
# Check for SPACE_HOST and SPACE_ID at startup for information
|
| 177 |
-
space_host_startup = os.getenv("SPACE_HOST")
|
| 178 |
-
space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
|
| 179 |
-
|
| 180 |
-
if space_host_startup:
|
| 181 |
-
print(f"✅ SPACE_HOST found: {space_host_startup}")
|
| 182 |
-
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
|
| 183 |
-
else:
|
| 184 |
-
print("ℹ️ SPACE_HOST environment variable not found (running locally?).")
|
| 185 |
-
|
| 186 |
-
if space_id_startup: # Print repo URLs if SPACE_ID is found
|
| 187 |
-
print(f"✅ SPACE_ID found: {space_id_startup}")
|
| 188 |
-
print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}")
|
| 189 |
-
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
|
| 190 |
-
else:
|
| 191 |
-
print("ℹ️ SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")
|
| 192 |
-
|
| 193 |
-
print("-"*(60 + len(" App Starting ")) + "\n")
|
| 194 |
-
|
| 195 |
-
print("Launching Gradio Interface for Basic Agent Evaluation...")
|
| 196 |
-
demo.launch(debug=True, share=False)
|
|
|
|
| 1 |
import os
|
| 2 |
+
import tempfile
|
| 3 |
import requests
|
| 4 |
+
from typing import Dict, Any, Annotated
|
| 5 |
+
from typing_extensions import TypedDict
|
| 6 |
+
import gradio as gr
|
| 7 |
import pandas as pd
|
| 8 |
|
| 9 |
+
# Your constants + imports stay
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
+
# New imports for the stack
|
| 12 |
+
from smolagents import CodeAgent, HfApiModel # Smolagents for code/web agents
|
| 13 |
+
from smolagents.tools import DuckDuckGoSearchResults # Built-in web tool
|
| 14 |
+
from langgraph.graph import StateGraph, END
|
| 15 |
+
from langgraph.prebuilt import ToolNode, tools_condition
|
| 16 |
+
from langchain_core.tools import tool
|
| 17 |
+
from langchain_core.messages import HumanMessage, AIMessage
|
| 18 |
+
from transformers import pipeline # For lightweight LLM routing
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
+
# --- Enhanced Agent with LangGraph + Smolagents ---
|
| 21 |
+
class CrmAgent:
|
| 22 |
+
def __init__(self):
|
| 23 |
+
print("CrmAgent initialized with LangGraph + Smolagents.")
|
| 24 |
+
# Lightweight router LLM (free HF inference)
|
| 25 |
+
self.router = pipeline("text-generation", model="gpt2", device=-1) # CPU for hack
|
| 26 |
+
# Smolagents CodeAgent with web tool
|
| 27 |
+
self.llm = HfApiModel(model_id="microsoft/DialoGPT-medium") # Free HF model
|
| 28 |
+
search_tool = DuckDuckGoSearchResults(num_results=3) # Quick web hits
|
| 29 |
+
self.code_agent = CodeAgent(llm=self.llm, tools=[search_tool])
|
| 30 |
+
# Temp dir for files
|
| 31 |
+
self.temp_dir = tempfile.mkdtemp()
|
| 32 |
+
|
| 33 |
+
# Tool: Download file if needed (GAIA questions may have attachments)
|
| 34 |
+
@tool
|
| 35 |
+
def download_file(self, task_id: str) -> str:
|
| 36 |
+
"""Downloads file for task_id if exists, returns path."""
|
| 37 |
+
url = f"{DEFAULT_API_URL}/files/{task_id}"
|
| 38 |
+
try:
|
| 39 |
+
resp = requests.get(url, timeout=10)
|
| 40 |
+
if resp.status_code == 200:
|
| 41 |
+
file_path = os.path.join(self.temp_dir, f"{task_id}_file")
|
| 42 |
+
with open(file_path, "wb") as f:
|
| 43 |
+
f.write(resp.content)
|
| 44 |
+
return f"File downloaded: {file_path}"
|
| 45 |
+
return "No file found."
|
| 46 |
+
except Exception as e:
|
| 47 |
+
return f"Download error: {e}"
|
| 48 |
+
|
| 49 |
+
# Router Node: Decide path with LLM
|
| 50 |
+
def router_node(self, state: Dict[str, Any]) -> Dict[str, str]:
|
| 51 |
+
question = state["question"]
|
| 52 |
+
prompt = f"Given question: '{question[:100]}...'. Respond with route: 'search' if needs web info, 'code' if math/file/code, 'both' if both, 'direct' if obvious."
|
| 53 |
+
response = self.router(prompt, max_length=20, num_return_sequences=1)[0]["generated_text"]
|
| 54 |
+
route = response.strip().lower().split()[-1] # Crude parse, tweak as needed
|
| 55 |
+
state["route"] = route
|
| 56 |
+
print(f"Routed to: {route}")
|
| 57 |
+
return state
|
| 58 |
+
|
| 59 |
+
# Search Node: Use smolagents web
|
| 60 |
+
def search_node(self, state: Dict[str, Any]) -> Dict[str, Any]:
|
| 61 |
+
question = state["question"]
|
| 62 |
+
try:
|
| 63 |
+
# Smolagents call (it handles tool selection internally)
|
| 64 |
+
result = self.code_agent.run(question) # Runs code/web as needed
|
| 65 |
+
state["search_results"] = result
|
| 66 |
+
print(f"Search/code output: {result[:100]}...")
|
| 67 |
+
except Exception as e:
|
| 68 |
+
state["search_results"] = f"Error: {e}"
|
| 69 |
+
return state
|
| 70 |
+
|
| 71 |
+
# Direct Node: Simple guess or pass
|
| 72 |
+
def direct_node(self, state: Dict[str, Any]) -> Dict[str, Any]:
|
| 73 |
+
# Fallback: Basic heuristic or empty
|
| 74 |
+
state["final_answer"] = "Direct answer needed—implement heuristic here."
|
| 75 |
+
return state
|
| 76 |
+
|
| 77 |
+
# Conditional Edge: Based on route
|
| 78 |
+
def conditional_route(self, state: Dict[str, Any]) -> str:
|
| 79 |
+
route = state.get("route", "direct")
|
| 80 |
+
if route in ["search", "both"]:
|
| 81 |
+
return "search"
|
| 82 |
+
elif route == "code":
|
| 83 |
+
return "search" # Smolagents handles code too
|
| 84 |
+
return "direct"
|
| 85 |
+
|
| 86 |
+
# Build the Graph
|
| 87 |
+
def build_graph(self):
|
| 88 |
+
# State
|
| 89 |
+
class AgentState(TypedDict):
|
| 90 |
+
question: str
|
| 91 |
+
route: str
|
| 92 |
+
search_results: str
|
| 93 |
+
final_answer: str
|
| 94 |
+
|
| 95 |
+
# Graph
|
| 96 |
+
workflow = StateGraph(AgentState)
|
| 97 |
+
workflow.add_node("router", self.router_node)
|
| 98 |
+
workflow.add_node("search", self.search_node)
|
| 99 |
+
workflow.add_node("direct", self.direct_node)
|
| 100 |
+
|
| 101 |
+
# Edges
|
| 102 |
+
workflow.set_entry_point("router")
|
| 103 |
+
workflow.add_conditional_edges("router", self.conditional_route, {"search": "search", "direct": "direct"})
|
| 104 |
+
workflow.add_edge("search", END)
|
| 105 |
+
workflow.add_edge("direct", END)
|
| 106 |
+
|
| 107 |
+
# Compile
|
| 108 |
+
self.graph = workflow.compile()
|
| 109 |
+
|
| 110 |
+
def __call__(self, question: str, task_id: str = None) -> str:
|
| 111 |
+
if not hasattr(self, "graph"):
|
| 112 |
+
self.build_graph()
|
| 113 |
+
# Download file if task_id
|
| 114 |
+
if task_id:
|
| 115 |
+
file_info = self.download_file.invoke({"task_id": task_id})
|
| 116 |
+
question += f" [File info: {file_info}]" # Append to prompt
|
| 117 |
+
|
| 118 |
+
# Run graph
|
| 119 |
+
initial_state = {"question": question, "route": "", "search_results": "", "final_answer": ""}
|
| 120 |
+
final_state = self.graph.invoke(initial_state)
|
| 121 |
+
|
| 122 |
+
# Extract clean answer (smolagents outputs code-thought → result)
|
| 123 |
+
answer = final_state.get("search_results", final_state.get("final_answer", "No answer generated."))
|
| 124 |
+
# Strip to exact (no extras)
|
| 125 |
+
if "final answer" in answer.lower():
|
| 126 |
+
answer = answer.split("final answer")[-1].strip().split()[0] if answer.split("final answer")[-1].strip() else answer
|
| 127 |
+
print(f"Agent final: {answer}")
|
| 128 |
+
return answer
|
| 129 |
+
|
| 130 |
+
# --- Update run_and_submit_all (minor tweak for task_id) ---
|
| 131 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 132 |
+
# ... (keep all your existing code up to agent init)
|
| 133 |
+
|
| 134 |
+
# 1. Instantiate Agent
|
| 135 |
try:
|
| 136 |
+
agent = CrmAgent() # Our new beast
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 137 |
except Exception as e:
|
| 138 |
+
# ...
|
| 139 |
+
|
| 140 |
+
# 3. Run your Agent (pass task_id)
|
|
|
|
| 141 |
results_log = []
|
| 142 |
answers_payload = []
|
| 143 |
print(f"Running agent on {len(questions_data)} questions...")
|
|
|
|
| 145 |
task_id = item.get("task_id")
|
| 146 |
question_text = item.get("question")
|
| 147 |
if not task_id or question_text is None:
|
| 148 |
+
# ...
|
|
|
|
| 149 |
try:
|
| 150 |
+
submitted_answer = agent(question_text, task_id) # Pass task_id for files
|
| 151 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 152 |
+
results_log.append({"Task ID": task_id, "Question": question_text[:50] + "...", "Submitted Answer": submitted_answer})
|
| 153 |
except Exception as e:
|
| 154 |
+
# ...
|
| 155 |
+
|
| 156 |
+
# ... (rest unchanged—submit as before)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|