Update app.py
Browse files
app.py
CHANGED
|
@@ -3,45 +3,88 @@ import time
|
|
| 3 |
import requests
|
| 4 |
import gradio as gr
|
| 5 |
import pandas as pd
|
| 6 |
-
from
|
| 7 |
|
| 8 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 9 |
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
class BasicAgent:
|
| 12 |
def __init__(self):
|
| 13 |
-
print("Initializing
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
model_id="Qwen/Qwen2.5-Coder-32B-Instruct",
|
| 17 |
token=os.environ.get("HF_TOKEN"),
|
| 18 |
)
|
| 19 |
-
|
| 20 |
-
self.agent = CodeAgent(
|
| 21 |
-
model=self.model,
|
| 22 |
-
tools=[DuckDuckGoSearchTool(), VisitWebpageTool()],
|
| 23 |
-
max_steps=3,
|
| 24 |
-
verbosity_level=0,
|
| 25 |
-
)
|
| 26 |
print("✅ Ready")
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
def __call__(self, question: str, task_id: str = None) -> str:
|
| 29 |
# Handle reversed text
|
| 30 |
-
if '.rewsna' in question or 'tfel' in question:
|
| 31 |
question = question[::-1]
|
|
|
|
| 32 |
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
return "unknown"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
|
| 46 |
|
| 47 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
@@ -51,7 +94,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 51 |
username = profile.username
|
| 52 |
space_id = os.getenv("SPACE_ID")
|
| 53 |
|
| 54 |
-
print(f"\nUser: {username}")
|
| 55 |
|
| 56 |
try:
|
| 57 |
agent = BasicAgent()
|
|
@@ -73,11 +116,20 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 73 |
question = q.get("question", "")
|
| 74 |
|
| 75 |
print(f"[{i+1}] {question[:50]}...")
|
| 76 |
-
|
| 77 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
answers.append({"task_id": task_id, "submitted_answer": answer})
|
| 80 |
results.append({"#": i+1, "Q": question[:40]+"...", "A": answer[:50]})
|
|
|
|
|
|
|
|
|
|
| 81 |
|
| 82 |
total = time.time() - start
|
| 83 |
print(f"\n⏱️ {total:.0f}s")
|
|
@@ -85,7 +137,11 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 85 |
try:
|
| 86 |
result = requests.post(
|
| 87 |
f"{DEFAULT_API_URL}/submit",
|
| 88 |
-
json={
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
timeout=60
|
| 90 |
).json()
|
| 91 |
|
|
@@ -101,7 +157,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 101 |
|
| 102 |
|
| 103 |
with gr.Blocks() as demo:
|
| 104 |
-
gr.Markdown("# 🎯 GAIA Agent -
|
|
|
|
| 105 |
gr.LoginButton()
|
| 106 |
btn = gr.Button("🚀 Run", variant="primary")
|
| 107 |
status = gr.Textbox(label="Status", lines=5)
|
|
|
|
| 3 |
import requests
|
| 4 |
import gradio as gr
|
| 5 |
import pandas as pd
|
| 6 |
+
from huggingface_hub import InferenceClient
|
| 7 |
|
| 8 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 9 |
|
| 10 |
|
| 11 |
+
def web_search(query: str) -> str:
|
| 12 |
+
"""Search using DuckDuckGo"""
|
| 13 |
+
try:
|
| 14 |
+
from duckduckgo_search import DDGS
|
| 15 |
+
with DDGS() as ddgs:
|
| 16 |
+
results = list(ddgs.text(query, max_results=3))
|
| 17 |
+
if results:
|
| 18 |
+
return "\n".join([f"- {r['title']}: {r['body']}" for r in results])
|
| 19 |
+
except:
|
| 20 |
+
pass
|
| 21 |
+
return ""
|
| 22 |
+
|
| 23 |
+
|
| 24 |
class BasicAgent:
|
| 25 |
def __init__(self):
|
| 26 |
+
print("Initializing agent...")
|
| 27 |
+
self.client = InferenceClient(
|
| 28 |
+
model="Qwen/Qwen2.5-72B-Instruct",
|
|
|
|
| 29 |
token=os.environ.get("HF_TOKEN"),
|
| 30 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
print("✅ Ready")
|
| 32 |
|
| 33 |
+
def ask(self, prompt: str) -> str:
|
| 34 |
+
"""Simple LLM call"""
|
| 35 |
+
try:
|
| 36 |
+
response = self.client.chat_completion(
|
| 37 |
+
messages=[{"role": "user", "content": prompt}],
|
| 38 |
+
max_tokens=50,
|
| 39 |
+
temperature=0.1,
|
| 40 |
+
)
|
| 41 |
+
return response.choices[0].message.content.strip()
|
| 42 |
+
except Exception as e:
|
| 43 |
+
print(f" LLM error: {e}")
|
| 44 |
+
return ""
|
| 45 |
+
|
| 46 |
def __call__(self, question: str, task_id: str = None) -> str:
|
| 47 |
# Handle reversed text
|
| 48 |
+
if '.rewsna' in question or 'tfel' in question or 'eht fo' in question:
|
| 49 |
question = question[::-1]
|
| 50 |
+
print(f" [Reversed → {question[:50]}...]")
|
| 51 |
|
| 52 |
+
# Search for context
|
| 53 |
+
search_results = web_search(question[:100])
|
| 54 |
+
|
| 55 |
+
# Build simple prompt
|
| 56 |
+
context = f"Search results:\n{search_results}\n\n" if search_results else ""
|
| 57 |
+
|
| 58 |
+
prompt = f"""{context}Question: {question}
|
| 59 |
+
|
| 60 |
+
Answer with ONLY the final answer.
|
| 61 |
+
- If it's a number, just the number (e.g., "42")
|
| 62 |
+
- If it's a name, just the name (e.g., "John Smith")
|
| 63 |
+
- If it's a list, comma-separated (e.g., "apple, banana, cherry")
|
| 64 |
+
- Maximum 5 words
|
| 65 |
+
|
| 66 |
+
Answer:"""
|
| 67 |
+
|
| 68 |
+
answer = self.ask(prompt)
|
| 69 |
+
|
| 70 |
+
# Clean the answer
|
| 71 |
+
if not answer:
|
| 72 |
return "unknown"
|
| 73 |
+
|
| 74 |
+
# Remove common prefixes
|
| 75 |
+
for prefix in ["Answer:", "The answer is:", "The answer is", "A:", "Final answer:"]:
|
| 76 |
+
if answer.lower().startswith(prefix.lower()):
|
| 77 |
+
answer = answer[len(prefix):].strip()
|
| 78 |
+
|
| 79 |
+
# Remove quotes and periods
|
| 80 |
+
answer = answer.strip('."\'')
|
| 81 |
+
|
| 82 |
+
# If answer is too long or contains excuses, retry with simpler prompt
|
| 83 |
+
if len(answer) > 100 or any(x in answer.lower() for x in ["i cannot", "i don't", "unable"]):
|
| 84 |
+
answer = self.ask(f"In 1-3 words, answer: {question}")
|
| 85 |
+
answer = answer.strip('."\'')
|
| 86 |
+
|
| 87 |
+
return answer if answer else "unknown"
|
| 88 |
|
| 89 |
|
| 90 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
|
|
| 94 |
username = profile.username
|
| 95 |
space_id = os.getenv("SPACE_ID")
|
| 96 |
|
| 97 |
+
print(f"\n{'='*40}\nUser: {username}\n{'='*40}")
|
| 98 |
|
| 99 |
try:
|
| 100 |
agent = BasicAgent()
|
|
|
|
| 116 |
question = q.get("question", "")
|
| 117 |
|
| 118 |
print(f"[{i+1}] {question[:50]}...")
|
| 119 |
+
|
| 120 |
+
try:
|
| 121 |
+
answer = agent(question, task_id)
|
| 122 |
+
except Exception as e:
|
| 123 |
+
print(f" Error: {e}")
|
| 124 |
+
answer = "unknown"
|
| 125 |
+
|
| 126 |
+
print(f" → {answer}")
|
| 127 |
|
| 128 |
answers.append({"task_id": task_id, "submitted_answer": answer})
|
| 129 |
results.append({"#": i+1, "Q": question[:40]+"...", "A": answer[:50]})
|
| 130 |
+
|
| 131 |
+
# Small delay to avoid rate limits
|
| 132 |
+
time.sleep(1)
|
| 133 |
|
| 134 |
total = time.time() - start
|
| 135 |
print(f"\n⏱️ {total:.0f}s")
|
|
|
|
| 137 |
try:
|
| 138 |
result = requests.post(
|
| 139 |
f"{DEFAULT_API_URL}/submit",
|
| 140 |
+
json={
|
| 141 |
+
"username": username,
|
| 142 |
+
"agent_code": f"https://huggingface.co/spaces/{space_id}/tree/main",
|
| 143 |
+
"answers": answers
|
| 144 |
+
},
|
| 145 |
timeout=60
|
| 146 |
).json()
|
| 147 |
|
|
|
|
| 157 |
|
| 158 |
|
| 159 |
with gr.Blocks() as demo:
|
| 160 |
+
gr.Markdown("# 🎯 GAIA Agent - Simple Mode")
|
| 161 |
+
gr.Markdown("Direct search + LLM (no code execution)")
|
| 162 |
gr.LoginButton()
|
| 163 |
btn = gr.Button("🚀 Run", variant="primary")
|
| 164 |
status = gr.Textbox(label="Status", lines=5)
|