Update app.py
Browse files
app.py
CHANGED
@@ -3,90 +3,113 @@ import gradio as gr
|
|
3 |
import requests
|
4 |
import pandas as pd
|
5 |
from io import BytesIO
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
# --- LangChain & Groq Imports ---
|
8 |
from groq import Groq
|
9 |
from langchain_groq import ChatGroq
|
10 |
from langchain.agents import AgentExecutor, create_tool_calling_agent
|
11 |
-
from
|
12 |
from langchain_core.prompts import ChatPromptTemplate
|
13 |
from langchain.tools import Tool
|
14 |
|
15 |
|
16 |
# --- Constants ---
|
17 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
|
|
18 |
|
19 |
|
20 |
-
# ---
|
21 |
-
def
|
22 |
"""
|
23 |
-
Downloads an audio file for a given task_id
|
24 |
-
|
25 |
-
Use this tool ONLY when a question explicitly mentions an audio file or recording.
|
26 |
-
The task_id MUST be provided as the input.
|
27 |
"""
|
28 |
-
print(f"Tool '
|
29 |
try:
|
30 |
-
# Step 1: Download the file
|
31 |
file_url = f"{DEFAULT_API_URL}/files/{task_id}"
|
32 |
-
print(f"Downloading audio file from: {file_url}")
|
33 |
audio_response = requests.get(file_url)
|
34 |
audio_response.raise_for_status()
|
35 |
-
|
36 |
-
# Step 2: Prepare the file for the Groq API
|
37 |
audio_bytes = BytesIO(audio_response.content)
|
38 |
-
audio_bytes.name = f"{task_id}.mp3"
|
39 |
-
|
40 |
-
# Step 3: Initialize the Groq client and transcribe
|
41 |
-
print("Initializing Groq client for transcription...")
|
42 |
client = Groq(api_key=os.getenv("GROQ_API_KEY"))
|
|
|
|
|
|
|
|
|
43 |
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
|
|
|
|
|
|
|
|
55 |
except Exception as e:
|
56 |
-
|
57 |
-
|
58 |
-
|
|
|
59 |
|
60 |
|
61 |
# --- Agent Definition ---
|
62 |
class LangChainAgent:
|
63 |
def __init__(self, groq_api_key: str, tavily_api_key: str):
|
64 |
-
print("Initializing LangChainAgent...")
|
65 |
-
|
66 |
-
# THIS IS THE CORRECTED LINE
|
67 |
self.llm = ChatGroq(model_name="llama3-70b-8192", groq_api_key=groq_api_key, temperature=0.0)
|
68 |
|
69 |
-
#
|
70 |
-
audio_tool = Tool(
|
71 |
-
name="audio_transcriber",
|
72 |
-
func=transcribe_audio_from_task_id,
|
73 |
-
description="Use this tool to transcribe an audio file. The input must be the task_id of the question.",
|
74 |
-
)
|
75 |
self.tools = [
|
76 |
-
TavilySearchResults(
|
77 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
]
|
79 |
|
80 |
-
#
|
81 |
prompt = ChatPromptTemplate.from_messages([
|
82 |
("system", (
|
83 |
"You are a powerful problem-solving agent. Your goal is to answer the user's question accurately. "
|
84 |
-
"You have access to
|
85 |
-
"
|
86 |
-
"
|
87 |
-
"
|
88 |
-
"-
|
89 |
-
"-
|
|
|
|
|
|
|
90 |
)),
|
91 |
("human", "Question: {input}\nTask ID: {task_id}"),
|
92 |
("placeholder", "{agent_scratchpad}"),
|
@@ -94,59 +117,52 @@ class LangChainAgent:
|
|
94 |
|
95 |
agent = create_tool_calling_agent(self.llm, self.tools, prompt)
|
96 |
self.agent_executor = AgentExecutor(agent=agent, tools=self.tools, verbose=True, handle_parsing_errors=True)
|
97 |
-
print("LangChainAgent initialized.")
|
98 |
|
99 |
def __call__(self, question: str, task_id: str) -> str:
|
100 |
-
|
|
|
|
|
|
|
|
|
|
|
101 |
try:
|
102 |
-
response = self.agent_executor.invoke(
|
103 |
-
|
104 |
except Exception as e:
|
105 |
-
|
106 |
-
print(f"Agent generated answer: {answer}")
|
107 |
-
return answer
|
108 |
-
|
109 |
|
110 |
-
# --- Main Application Logic ---
|
111 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
112 |
space_id = os.getenv("SPACE_ID")
|
113 |
-
if not profile:
|
114 |
-
return "Please Login to Hugging Face with the button.", None
|
115 |
username = profile.username
|
116 |
-
print(f"User logged in: {username}")
|
117 |
-
|
118 |
try:
|
119 |
groq_api_key = os.getenv("GROQ_API_KEY")
|
120 |
tavily_api_key = os.getenv("TAVILY_API_KEY")
|
121 |
-
if not all([groq_api_key, tavily_api_key]):
|
122 |
-
raise ValueError("An API key secret (GROQ or TAVILY) is missing.")
|
123 |
agent = LangChainAgent(groq_api_key=groq_api_key, tavily_api_key=tavily_api_key)
|
124 |
-
except Exception as e:
|
125 |
-
|
126 |
-
|
127 |
questions_url = f"{DEFAULT_API_URL}/questions"
|
128 |
-
print(f"Fetching questions from: {questions_url}")
|
129 |
try:
|
130 |
response = requests.get(questions_url, timeout=20)
|
131 |
response.raise_for_status()
|
132 |
questions_data = response.json()
|
133 |
-
except Exception as e:
|
134 |
-
|
135 |
-
|
136 |
results_log, answers_payload = [], []
|
137 |
for item in questions_data:
|
138 |
-
task_id,
|
139 |
-
if not task_id or not
|
140 |
-
|
141 |
-
answers_payload.append({"task_id": task_id, "submitted_answer":
|
142 |
-
results_log.append({"Task ID": task_id, "Question":
|
143 |
-
|
144 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
145 |
submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
|
146 |
submit_url = f"{DEFAULT_API_URL}/submit"
|
147 |
-
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
148 |
try:
|
149 |
-
response = requests.post(submit_url, json=submission_data, timeout=
|
150 |
response.raise_for_status()
|
151 |
result_data = response.json()
|
152 |
final_status = (f"Submission Successful!\nUser: {result_data.get('username')}\n"
|
@@ -154,14 +170,12 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
154 |
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
155 |
f"Message: {result_data.get('message', 'No message received.')}")
|
156 |
return final_status, pd.DataFrame(results_log)
|
157 |
-
except Exception as e:
|
158 |
-
return f"Submission Failed: {e}", pd.DataFrame(results_log)
|
159 |
-
|
160 |
|
161 |
-
# --- Gradio Interface ---
|
162 |
with gr.Blocks() as demo:
|
163 |
-
gr.Markdown("#
|
164 |
-
gr.Markdown("This agent can search
|
165 |
gr.LoginButton()
|
166 |
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
167 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
|
|
3 |
import requests
|
4 |
import pandas as pd
|
5 |
from io import BytesIO
|
6 |
+
import re
|
7 |
+
|
8 |
+
# --- New Imports for Video Tool ---
|
9 |
+
from pytube import YouTube
|
10 |
+
import moviepy.editor as mp
|
11 |
|
12 |
# --- LangChain & Groq Imports ---
|
13 |
from groq import Groq
|
14 |
from langchain_groq import ChatGroq
|
15 |
from langchain.agents import AgentExecutor, create_tool_calling_agent
|
16 |
+
from langchain_tavily_search import TavilySearchResults
|
17 |
from langchain_core.prompts import ChatPromptTemplate
|
18 |
from langchain.tools import Tool
|
19 |
|
20 |
|
21 |
# --- Constants ---
|
22 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
23 |
+
TEMP_DIR = "/tmp" # Use the /tmp directory for temporary files in HF Spaces
|
24 |
|
25 |
|
26 |
+
# --- Tool Definition: Audio File Transcription ---
|
27 |
+
def transcribe_audio_file(task_id: str) -> str:
|
28 |
"""
|
29 |
+
Downloads an audio file (.mp3) for a given task_id, transcribes it, and returns the text.
|
30 |
+
Use this tool ONLY when a question explicitly mentions an audio file, .mp3, recording, or voice memo.
|
|
|
|
|
31 |
"""
|
32 |
+
print(f"Tool 'transcribe_audio_file' called with task_id: {task_id}")
|
33 |
try:
|
|
|
34 |
file_url = f"{DEFAULT_API_URL}/files/{task_id}"
|
|
|
35 |
audio_response = requests.get(file_url)
|
36 |
audio_response.raise_for_status()
|
|
|
|
|
37 |
audio_bytes = BytesIO(audio_response.content)
|
38 |
+
audio_bytes.name = f"{task_id}.mp3"
|
39 |
+
|
|
|
|
|
40 |
client = Groq(api_key=os.getenv("GROQ_API_KEY"))
|
41 |
+
transcription = client.audio.transcriptions.create(file=audio_bytes, model="whisper-large-v3", response_format="text")
|
42 |
+
return str(transcription)
|
43 |
+
except Exception as e:
|
44 |
+
return f"Error during audio file transcription: {e}"
|
45 |
|
46 |
+
# --- Tool Definition: Video Transcription ---
|
47 |
+
def transcribe_youtube_video(video_url: str) -> str:
|
48 |
+
"""
|
49 |
+
Downloads a YouTube video from a URL, extracts its audio, and transcribes it to text.
|
50 |
+
Use this tool ONLY when a question provides a youtube.com URL.
|
51 |
+
"""
|
52 |
+
print(f"Tool 'transcribe_youtube_video' called with URL: {video_url}")
|
53 |
+
video_path, audio_path = None, None
|
54 |
+
try:
|
55 |
+
os.makedirs(TEMP_DIR, exist_ok=True)
|
56 |
+
yt = YouTube(video_url)
|
57 |
+
stream = yt.streams.filter(progressive=True, file_extension='mp4').order_by('resolution').desc().first()
|
58 |
+
video_path = stream.download(output_path=TEMP_DIR)
|
59 |
|
60 |
+
video_clip = mp.VideoFileClip(video_path)
|
61 |
+
audio_path = os.path.join(TEMP_DIR, "temp_audio.mp3")
|
62 |
+
video_clip.audio.write_audiofile(audio_path, codec='mp3', logger=None)
|
63 |
+
|
64 |
+
client = Groq(api_key=os.getenv("GROQ_API_KEY"))
|
65 |
+
with open(audio_path, "rb") as audio_file:
|
66 |
+
transcription = client.audio.transcriptions.create(file=audio_file, model="whisper-large-v3", response_format="text")
|
67 |
+
return str(transcription)
|
68 |
except Exception as e:
|
69 |
+
return f"Error during YouTube transcription: {e}"
|
70 |
+
finally:
|
71 |
+
if video_path and os.path.exists(video_path): os.remove(video_path)
|
72 |
+
if audio_path and os.path.exists(audio_path): os.remove(audio_path)
|
73 |
|
74 |
|
75 |
# --- Agent Definition ---
|
76 |
class LangChainAgent:
|
77 |
def __init__(self, groq_api_key: str, tavily_api_key: str):
|
|
|
|
|
|
|
78 |
self.llm = ChatGroq(model_name="llama3-70b-8192", groq_api_key=groq_api_key, temperature=0.0)
|
79 |
|
80 |
+
# Updated tools with much more specific descriptions
|
|
|
|
|
|
|
|
|
|
|
81 |
self.tools = [
|
82 |
+
TavilySearchResults(
|
83 |
+
name="web_search",
|
84 |
+
max_results=3,
|
85 |
+
tavily_api_key=tavily_api_key,
|
86 |
+
description="A search engine for finding up-to-date information, facts, and news on the internet."
|
87 |
+
),
|
88 |
+
Tool(
|
89 |
+
name="audio_file_transcriber",
|
90 |
+
func=transcribe_audio_file,
|
91 |
+
description="Use this ONLY for questions mentioning an audio file (.mp3, recording). Input MUST be the task_id.",
|
92 |
+
),
|
93 |
+
Tool(
|
94 |
+
name="youtube_video_transcriber",
|
95 |
+
func=transcribe_youtube_video,
|
96 |
+
description="Use this ONLY for questions providing a youtube.com URL. Input MUST be the URL.",
|
97 |
+
),
|
98 |
]
|
99 |
|
100 |
+
# Updated, rule-based system prompt
|
101 |
prompt = ChatPromptTemplate.from_messages([
|
102 |
("system", (
|
103 |
"You are a powerful problem-solving agent. Your goal is to answer the user's question accurately. "
|
104 |
+
"You have access to a web search tool, an audio file transcriber, and a YouTube video transcriber.\n\n"
|
105 |
+
"**REASONING PROCESS:**\n"
|
106 |
+
"1. **Analyze the question:** Is it a general knowledge question, or does it mention a file/URL?\n"
|
107 |
+
"2. **Select ONE tool:**\n"
|
108 |
+
" - If the question requires current events, facts, or general knowledge, use `web_search`.\n"
|
109 |
+
" - If the question *explicitly* mentions an audio file, .mp3, or voice memo, use `audio_file_transcriber` with the provided `task_id`.\n"
|
110 |
+
" - If the question *explicitly* provides a `youtube.com` URL, use `youtube_video_transcriber` with that URL.\n"
|
111 |
+
" - If no tool is needed (e.g., math, logic puzzles), answer directly.\n"
|
112 |
+
"3. **Execute and Answer:** After using a tool, analyze the result and provide ONLY THE FINAL ANSWER. Do not explain your actions or apologize for errors."
|
113 |
)),
|
114 |
("human", "Question: {input}\nTask ID: {task_id}"),
|
115 |
("placeholder", "{agent_scratchpad}"),
|
|
|
117 |
|
118 |
agent = create_tool_calling_agent(self.llm, self.tools, prompt)
|
119 |
self.agent_executor = AgentExecutor(agent=agent, tools=self.tools, verbose=True, handle_parsing_errors=True)
|
|
|
120 |
|
121 |
def __call__(self, question: str, task_id: str) -> str:
|
122 |
+
# The agent sometimes needs the URL directly, so let's extract it if present.
|
123 |
+
urls = re.findall(r'https?://[^\s]+', question)
|
124 |
+
input_for_agent = {"input": question, "task_id": task_id}
|
125 |
+
if urls:
|
126 |
+
input_for_agent['video_url'] = urls[0] # Pass the URL if found
|
127 |
+
|
128 |
try:
|
129 |
+
response = self.agent_executor.invoke(input_for_agent)
|
130 |
+
return response.get("output", "Agent failed to produce an answer.")
|
131 |
except Exception as e:
|
132 |
+
return f"Agent execution failed with an error: {e}"
|
|
|
|
|
|
|
133 |
|
134 |
+
# --- Main Application Logic (Unchanged) ---
|
135 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
136 |
space_id = os.getenv("SPACE_ID")
|
137 |
+
if not profile: return "Please Login to Hugging Face with the button.", None
|
|
|
138 |
username = profile.username
|
|
|
|
|
139 |
try:
|
140 |
groq_api_key = os.getenv("GROQ_API_KEY")
|
141 |
tavily_api_key = os.getenv("TAVILY_API_KEY")
|
142 |
+
if not all([groq_api_key, tavily_api_key]): raise ValueError("GROQ or TAVILY API key is missing.")
|
|
|
143 |
agent = LangChainAgent(groq_api_key=groq_api_key, tavily_api_key=tavily_api_key)
|
144 |
+
except Exception as e: return f"Error initializing agent: {e}", None
|
145 |
+
|
|
|
146 |
questions_url = f"{DEFAULT_API_URL}/questions"
|
|
|
147 |
try:
|
148 |
response = requests.get(questions_url, timeout=20)
|
149 |
response.raise_for_status()
|
150 |
questions_data = response.json()
|
151 |
+
except Exception as e: return f"Error fetching questions: {e}", None
|
152 |
+
|
|
|
153 |
results_log, answers_payload = [], []
|
154 |
for item in questions_data:
|
155 |
+
task_id, q_text = item.get("task_id"), item.get("question")
|
156 |
+
if not task_id or not q_text: continue
|
157 |
+
answer = agent(question=q_text, task_id=task_id)
|
158 |
+
answers_payload.append({"task_id": task_id, "submitted_answer": answer})
|
159 |
+
results_log.append({"Task ID": task_id, "Question": q_text, "Submitted Answer": answer})
|
160 |
+
|
161 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
162 |
submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
|
163 |
submit_url = f"{DEFAULT_API_URL}/submit"
|
|
|
164 |
try:
|
165 |
+
response = requests.post(submit_url, json=submission_data, timeout=240) # Increased timeout for video processing
|
166 |
response.raise_for_status()
|
167 |
result_data = response.json()
|
168 |
final_status = (f"Submission Successful!\nUser: {result_data.get('username')}\n"
|
|
|
170 |
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
171 |
f"Message: {result_data.get('message', 'No message received.')}")
|
172 |
return final_status, pd.DataFrame(results_log)
|
173 |
+
except Exception as e: return f"Submission Failed: {e}", pd.DataFrame(results_log)
|
|
|
|
|
174 |
|
175 |
+
# --- Gradio Interface (Unchanged) ---
|
176 |
with gr.Blocks() as demo:
|
177 |
+
gr.Markdown("# Ultimate Agent Runner (Search + Audio + Video)")
|
178 |
+
gr.Markdown("This agent can search, transcribe audio files, and transcribe YouTube videos.")
|
179 |
gr.LoginButton()
|
180 |
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
181 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|