Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,8 +1,9 @@
|
|
|
|
|
| 1 |
"""
|
| 2 |
-
app.py
|
| 3 |
This script provides the Gradio web interface to run the evaluation.
|
| 4 |
-
This version is simplified to work with the new agent architecture
|
| 5 |
-
|
|
|
|
| 6 |
"""
|
| 7 |
|
| 8 |
import os
|
|
@@ -10,14 +11,16 @@ import re
|
|
| 10 |
import gradio as gr
|
| 11 |
import requests
|
| 12 |
import pandas as pd
|
| 13 |
-
|
|
|
|
| 14 |
|
| 15 |
from agent import create_agent_executor
|
| 16 |
|
| 17 |
# --- Constants ---
|
|
|
|
| 18 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 19 |
|
| 20 |
-
# --- Helper function to parse the agent's output ---
|
| 21 |
def parse_final_answer(agent_response: str) -> str:
|
| 22 |
match = re.search(r"FINAL ANSWER:\s*(.*)", agent_response, re.IGNORECASE | re.DOTALL)
|
| 23 |
if match: return match.group(1).strip()
|
|
@@ -25,6 +28,13 @@ def parse_final_answer(agent_response: str) -> str:
|
|
| 25 |
if lines: return lines[-1].strip()
|
| 26 |
return "Could not parse a final answer."
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 29 |
"""
|
| 30 |
Fetches all questions, runs the agent on them, submits all answers,
|
|
@@ -36,7 +46,17 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 36 |
username = profile.username
|
| 37 |
print(f"User logged in: {username}")
|
| 38 |
|
|
|
|
|
|
|
| 39 |
space_id = os.getenv("SPACE_ID")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 41 |
questions_url = f"{DEFAULT_API_URL}/questions"
|
| 42 |
submit_url = f"{DEFAULT_API_URL}/submit"
|
|
@@ -72,7 +92,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 72 |
|
| 73 |
file_url = item.get("file_url")
|
| 74 |
|
| 75 |
-
|
|
|
|
| 76 |
# The agent's multimodal_router will handle the rest.
|
| 77 |
if file_url:
|
| 78 |
full_question_text = f"{question_text}\n\nHere is the relevant file: {file_url}"
|
|
@@ -83,7 +104,11 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 83 |
print(f"Raw Prompt for Agent:\n{full_question_text}")
|
| 84 |
|
| 85 |
try:
|
| 86 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
raw_answer = result['messages'][-1].content
|
| 88 |
submitted_answer = parse_final_answer(raw_answer)
|
| 89 |
|
|
@@ -91,7 +116,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 91 |
print(f"PARSED FINAL ANSWER: '{submitted_answer}'")
|
| 92 |
|
| 93 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 94 |
-
# The log for the DataFrame no longer includes a 'File Type' column
|
| 95 |
results_log.append({
|
| 96 |
"Task ID": task_id,
|
| 97 |
"Question": question_text,
|
|
@@ -113,7 +137,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 113 |
if not answers_payload:
|
| 114 |
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 115 |
|
| 116 |
-
# 4. Prepare and 5. Submit
|
| 117 |
submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
|
| 118 |
print(f"\nSubmitting {len(answers_payload)} answers for user '{username}'...")
|
| 119 |
try:
|
|
@@ -129,31 +153,34 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 129 |
print(status_message)
|
| 130 |
return status_message, pd.DataFrame(results_log)
|
| 131 |
|
| 132 |
-
# --- Gradio UI ---
|
| 133 |
with gr.Blocks(title="Multimodal Agent Evaluation") as demo:
|
| 134 |
gr.Markdown("# Multimodal Agent Evaluation Runner")
|
| 135 |
gr.Markdown("This agent can process images, YouTube videos, audio files, and perform web searches.")
|
| 136 |
|
| 137 |
-
|
| 138 |
-
login_button = gr.LoginButton()
|
| 139 |
-
|
| 140 |
run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
|
| 141 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=6, interactive=False)
|
| 142 |
results_table = gr.DataFrame(
|
| 143 |
label="Questions and Agent Answers",
|
| 144 |
wrap=True,
|
| 145 |
row_count=10,
|
| 146 |
-
|
|
|
|
|
|
|
| 147 |
)
|
| 148 |
|
| 149 |
-
|
| 150 |
-
#
|
| 151 |
-
|
| 152 |
-
|
| 153 |
-
|
| 154 |
-
|
| 155 |
-
|
|
|
|
|
|
|
|
|
|
| 156 |
|
| 157 |
if __name__ == "__main__":
|
| 158 |
print("\n" + "-"*30 + " Multimodal App Starting " + "-"*30)
|
| 159 |
-
demo.launch()
|
|
|
|
| 1 |
+
# app.py
|
| 2 |
"""
|
|
|
|
| 3 |
This script provides the Gradio web interface to run the evaluation.
|
| 4 |
+
## MODIFICATION: This version is simplified to work with the new agent architecture.
|
| 5 |
+
It no longer performs file-type detection or prompt enhancement, as that responsibility
|
| 6 |
+
has been moved into the agent's 'multimodal_router'.
|
| 7 |
"""
|
| 8 |
|
| 9 |
import os
|
|
|
|
| 11 |
import gradio as gr
|
| 12 |
import requests
|
| 13 |
import pandas as pd
|
| 14 |
+
# --- Import HumanMessage ---
|
| 15 |
+
from langchain_core.messages import HumanMessage
|
| 16 |
|
| 17 |
from agent import create_agent_executor
|
| 18 |
|
| 19 |
# --- Constants ---
|
| 20 |
+
# Ensure the URL is correctly formatted (remove trailing spaces)
|
| 21 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 22 |
|
| 23 |
+
# --- Helper function to parse the agent's output (remains the same) ---
|
| 24 |
def parse_final_answer(agent_response: str) -> str:
|
| 25 |
match = re.search(r"FINAL ANSWER:\s*(.*)", agent_response, re.IGNORECASE | re.DOTALL)
|
| 26 |
if match: return match.group(1).strip()
|
|
|
|
| 28 |
if lines: return lines[-1].strip()
|
| 29 |
return "Could not parse a final answer."
|
| 30 |
|
| 31 |
+
## MODIFICATION: The `detect_file_type` function has been removed.
|
| 32 |
+
## It is now redundant as this logic is handled inside the agent.
|
| 33 |
+
|
| 34 |
+
## MODIFICATION: The `create_enhanced_prompt` function has been removed.
|
| 35 |
+
## It was causing errors by trying to instruct the agent to use tools that no longer exist.
|
| 36 |
+
## The agent is now responsible for handling the raw input itself.
|
| 37 |
+
|
| 38 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 39 |
"""
|
| 40 |
Fetches all questions, runs the agent on them, submits all answers,
|
|
|
|
| 46 |
username = profile.username
|
| 47 |
print(f"User logged in: {username}")
|
| 48 |
|
| 49 |
+
# --- Fix SPACE_ID retrieval and URL construction ---
|
| 50 |
+
# Ensure SPACE_ID environment variable is set correctly in your Hugging Face Space.
|
| 51 |
space_id = os.getenv("SPACE_ID")
|
| 52 |
+
if not space_id:
|
| 53 |
+
# Fallback or error handling if SPACE_ID is not set
|
| 54 |
+
# You might need to adjust this based on how your space is configured
|
| 55 |
+
# For example, if running locally, you might not have SPACE_ID.
|
| 56 |
+
# This is a placeholder; adjust as needed.
|
| 57 |
+
# Consider using a default or making it configurable.
|
| 58 |
+
space_id = "your-username/your-space-name" # Example placeholder
|
| 59 |
+
print(f"Warning: SPACE_ID environment variable not found. Using placeholder: {space_id}")
|
| 60 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 61 |
questions_url = f"{DEFAULT_API_URL}/questions"
|
| 62 |
submit_url = f"{DEFAULT_API_URL}/submit"
|
|
|
|
| 92 |
|
| 93 |
file_url = item.get("file_url")
|
| 94 |
|
| 95 |
+
## MODIFICATION: Prompt creation is now much simpler.
|
| 96 |
+
# We just combine the question and the URL into one string.
|
| 97 |
# The agent's multimodal_router will handle the rest.
|
| 98 |
if file_url:
|
| 99 |
full_question_text = f"{question_text}\n\nHere is the relevant file: {file_url}"
|
|
|
|
| 104 |
print(f"Raw Prompt for Agent:\n{full_question_text}")
|
| 105 |
|
| 106 |
try:
|
| 107 |
+
# --- FIX: Pass a list of HumanMessage objects ---
|
| 108 |
+
# The agent expects MessagesState["messages"] to be a list of BaseMessage objects.
|
| 109 |
+
input_state = {"messages": [HumanMessage(content=full_question_text)]}
|
| 110 |
+
result = agent_executor.invoke(input_state)
|
| 111 |
+
|
| 112 |
raw_answer = result['messages'][-1].content
|
| 113 |
submitted_answer = parse_final_answer(raw_answer)
|
| 114 |
|
|
|
|
| 116 |
print(f"PARSED FINAL ANSWER: '{submitted_answer}'")
|
| 117 |
|
| 118 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
|
|
|
| 119 |
results_log.append({
|
| 120 |
"Task ID": task_id,
|
| 121 |
"Question": question_text,
|
|
|
|
| 137 |
if not answers_payload:
|
| 138 |
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 139 |
|
| 140 |
+
# 4. Prepare and 5. Submit (remains the same)
|
| 141 |
submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
|
| 142 |
print(f"\nSubmitting {len(answers_payload)} answers for user '{username}'...")
|
| 143 |
try:
|
|
|
|
| 153 |
print(status_message)
|
| 154 |
return status_message, pd.DataFrame(results_log)
|
| 155 |
|
| 156 |
+
# --- Gradio UI (remains largely the same) ---
|
| 157 |
with gr.Blocks(title="Multimodal Agent Evaluation") as demo:
|
| 158 |
gr.Markdown("# Multimodal Agent Evaluation Runner")
|
| 159 |
gr.Markdown("This agent can process images, YouTube videos, audio files, and perform web searches.")
|
| 160 |
|
| 161 |
+
gr.LoginButton()
|
|
|
|
|
|
|
| 162 |
run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
|
| 163 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=6, interactive=False)
|
| 164 |
results_table = gr.DataFrame(
|
| 165 |
label="Questions and Agent Answers",
|
| 166 |
wrap=True,
|
| 167 |
row_count=10,
|
| 168 |
+
# MODIFICATION: Removed the 'File Type' column as it's no longer detected here.
|
| 169 |
+
# Adjust column widths if necessary based on actual content/columns
|
| 170 |
+
# column_widths=[80, 250, 200, 250]
|
| 171 |
)
|
| 172 |
|
| 173 |
+
# We also remove "File Type" from the results_log being displayed
|
| 174 |
+
# (Though it's not in the log anymore, this is a safe check)
|
| 175 |
+
def display_wrapper(profile):
|
| 176 |
+
status, df = run_and_submit_all(profile)
|
| 177 |
+
# Ensure df is a DataFrame before attempting operations
|
| 178 |
+
if isinstance(df, pd.DataFrame) and "File Type" in df.columns:
|
| 179 |
+
df = df.drop(columns=["File Type"])
|
| 180 |
+
return status, df
|
| 181 |
+
|
| 182 |
+
run_button.click(fn=display_wrapper, outputs=[status_output, results_table])
|
| 183 |
|
| 184 |
if __name__ == "__main__":
|
| 185 |
print("\n" + "-"*30 + " Multimodal App Starting " + "-"*30)
|
| 186 |
+
demo.launch()
|