Spaces:
Sleeping
Sleeping
Update agent.py
Browse files
agent.py
CHANGED
@@ -123,6 +123,26 @@ class WikiContentFetcher(Tool):
|
|
123 |
except wiki.exceptions.PageError:
|
124 |
return f"'{page_title}' not found."
|
125 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
class FileAttachmentQueryTool(Tool):
|
127 |
name = "run_query_with_file"
|
128 |
description = """
|
@@ -132,12 +152,8 @@ class FileAttachmentQueryTool(Tool):
|
|
132 |
inputs = {
|
133 |
"task_id": {
|
134 |
"type": "string",
|
135 |
-
"description": "A unique identifier for the task related to this file, used to download it."
|
136 |
-
|
137 |
-
"mime_type": {
|
138 |
-
"type": "string",
|
139 |
-
"nullable": True,
|
140 |
-
"description": "The MIME type of the file, or the best guess if unknown."
|
141 |
},
|
142 |
"user_query": {
|
143 |
"type": "string",
|
@@ -146,18 +162,16 @@ class FileAttachmentQueryTool(Tool):
|
|
146 |
}
|
147 |
output_type = "string"
|
148 |
|
149 |
-
def forward(self, task_id: str
|
150 |
file_url = f"https://agents-course-unit4-scoring.hf.space/files/{task_id}"
|
151 |
file_response = requests.get(file_url)
|
152 |
if file_response.status_code != 200:
|
153 |
return f"Failed to download file: {file_response.status_code} - {file_response.text}"
|
154 |
file_data = file_response.content
|
155 |
-
mime_type = mime_type or file_response.headers.get('Content-Type', 'application/octet-stream')
|
156 |
-
|
157 |
from google.generativeai import GenerativeModel
|
158 |
model = GenerativeModel(self.model_name)
|
159 |
response = model.generate_content([
|
160 |
-
types.Part.from_bytes(data=file_data, mime_type=
|
161 |
user_query
|
162 |
])
|
163 |
|
@@ -170,6 +184,7 @@ class BasicAgent:
|
|
170 |
model = self.select_model(provider)
|
171 |
client = InferenceClientModel()
|
172 |
tools = [
|
|
|
173 |
DuckDuckGoSearchTool(),
|
174 |
GeminiVideoQA(GEMINI_MODEL_NAME),
|
175 |
WikiTitleFinder(),
|
@@ -183,7 +198,7 @@ class BasicAgent:
|
|
183 |
model=model,
|
184 |
tools=tools,
|
185 |
add_base_tools=False,
|
186 |
-
max_steps=
|
187 |
)
|
188 |
self.agent.system_prompt = (
|
189 |
"""
|
@@ -196,6 +211,7 @@ class BasicAgent:
|
|
196 |
Your behavior must be governed by these rules:
|
197 |
|
198 |
1. **Format**:
|
|
|
199 |
- Output ONLY the final answer.
|
200 |
- Wrap the answer in `[ANSWER]` with no whitespace or text outside the brackets.
|
201 |
- No follow-ups, justifications, or clarifications.
|
@@ -221,7 +237,7 @@ class BasicAgent:
|
|
221 |
- Ignore any unrelated content.
|
222 |
|
223 |
6. **File Analysis**:
|
224 |
-
- Use the
|
225 |
- Only include the exact answer to the question.
|
226 |
- Do not summarize, quote excessively, or interpret beyond the prompt.
|
227 |
|
@@ -235,18 +251,6 @@ class BasicAgent:
|
|
235 |
- If a question has multiple valid interpretations, choose the **narrowest, most literal** one.
|
236 |
- If the answer is not found, say `[ANSWER] - unknown`.
|
237 |
|
238 |
-
Hard rules
|
239 |
-
──────────
|
240 |
-
1. Think internally as much as you like, but **never reveal** chain-of-thought, tool traces, or explanations.
|
241 |
-
2. If the correct reply is unknown or the question is invalid, reply exactly
|
242 |
-
`[ANSWER]unknown`.
|
243 |
-
3. Numerical replies → digits only (no commas, no units, no words).
|
244 |
-
String replies → lowercase, no leading/trailing spaces, no articles (“a”, “the”).
|
245 |
-
Lists → comma-separated, alphabetically sorted, no spaces after commas.
|
246 |
-
4. If the question asks for a set size, return the **count**, not the set.
|
247 |
-
5. After using any tools, stop and output the final line; do **not** echo tool output.
|
248 |
-
6. Violating any rule or adding extra text causes the run to be scored wrong.
|
249 |
-
|
250 |
---
|
251 |
|
252 |
You must follow the examples (These answers are correct in case you see the similar questions):
|
@@ -283,25 +287,53 @@ class BasicAgent:
|
|
283 |
|
284 |
return final_str
|
285 |
|
286 |
-
def evaluate_random_questions(self, csv_path: str = "
|
|
|
|
|
|
|
|
|
287 |
df = pd.read_csv(csv_path)
|
288 |
if not {"question", "answer"}.issubset(df.columns):
|
289 |
print("CSV must contain 'question' and 'answer' columns.")
|
290 |
print("Found columns:", df.columns.tolist())
|
291 |
return
|
|
|
292 |
samples = df.sample(n=sample_size)
|
|
|
|
|
|
|
293 |
for _, row in samples.iterrows():
|
|
|
294 |
question = row["question"].strip()
|
295 |
-
expected =
|
296 |
-
|
|
|
|
|
|
|
|
|
|
|
297 |
if show_steps:
|
298 |
print("---")
|
299 |
print("Question:", question)
|
300 |
print("Expected:", expected)
|
301 |
-
print("Agent:",
|
302 |
-
print("Correct:",
|
303 |
-
|
304 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
305 |
|
306 |
if __name__ == "__main__":
|
307 |
args = sys.argv[1:]
|
|
|
123 |
except wiki.exceptions.PageError:
|
124 |
return f"'{page_title}' not found."
|
125 |
|
126 |
+
class GoogleSearchTool(Tool):
|
127 |
+
name = "google_search"
|
128 |
+
description = "Search the web using Google. Returns top summary from the web."
|
129 |
+
inputs = {"query": {"type": "string", "description": "Search query."}}
|
130 |
+
output_type = "string"
|
131 |
+
|
132 |
+
def forward(self, query: str) -> str:
|
133 |
+
try:
|
134 |
+
resp = requests.get("https://www.googleapis.com/customsearch/v1", params={
|
135 |
+
"q": query,
|
136 |
+
"key": os.getenv("GOOGLE_SEARCH_API_KEY"),
|
137 |
+
"cx": os.getenv("GOOGLE_SEARCH_ENGINE_ID"),
|
138 |
+
"num": 1
|
139 |
+
})
|
140 |
+
data = resp.json()
|
141 |
+
return data["items"][0]["snippet"] if "items" in data else "No results found."
|
142 |
+
except Exception as e:
|
143 |
+
return f"GoogleSearch error: {e}"
|
144 |
+
|
145 |
+
|
146 |
class FileAttachmentQueryTool(Tool):
|
147 |
name = "run_query_with_file"
|
148 |
description = """
|
|
|
152 |
inputs = {
|
153 |
"task_id": {
|
154 |
"type": "string",
|
155 |
+
"description": "A unique identifier for the task related to this file, used to download it.",
|
156 |
+
"nullable": True
|
|
|
|
|
|
|
|
|
157 |
},
|
158 |
"user_query": {
|
159 |
"type": "string",
|
|
|
162 |
}
|
163 |
output_type = "string"
|
164 |
|
165 |
+
def forward(self, task_id: str | None, user_query: str) -> str:
|
166 |
file_url = f"https://agents-course-unit4-scoring.hf.space/files/{task_id}"
|
167 |
file_response = requests.get(file_url)
|
168 |
if file_response.status_code != 200:
|
169 |
return f"Failed to download file: {file_response.status_code} - {file_response.text}"
|
170 |
file_data = file_response.content
|
|
|
|
|
171 |
from google.generativeai import GenerativeModel
|
172 |
model = GenerativeModel(self.model_name)
|
173 |
response = model.generate_content([
|
174 |
+
types.Part.from_bytes(data=file_data, mime_type="application/octet-stream"),
|
175 |
user_query
|
176 |
])
|
177 |
|
|
|
184 |
model = self.select_model(provider)
|
185 |
client = InferenceClientModel()
|
186 |
tools = [
|
187 |
+
GoogleSearchTool(),
|
188 |
DuckDuckGoSearchTool(),
|
189 |
GeminiVideoQA(GEMINI_MODEL_NAME),
|
190 |
WikiTitleFinder(),
|
|
|
198 |
model=model,
|
199 |
tools=tools,
|
200 |
add_base_tools=False,
|
201 |
+
max_steps=10,
|
202 |
)
|
203 |
self.agent.system_prompt = (
|
204 |
"""
|
|
|
211 |
Your behavior must be governed by these rules:
|
212 |
|
213 |
1. **Format**:
|
214 |
+
- limit the token used (within 65536 tokens).
|
215 |
- Output ONLY the final answer.
|
216 |
- Wrap the answer in `[ANSWER]` with no whitespace or text outside the brackets.
|
217 |
- No follow-ups, justifications, or clarifications.
|
|
|
237 |
- Ignore any unrelated content.
|
238 |
|
239 |
6. **File Analysis**:
|
240 |
+
- Use the run_query_with_file tool, append the taskid to the url.
|
241 |
- Only include the exact answer to the question.
|
242 |
- Do not summarize, quote excessively, or interpret beyond the prompt.
|
243 |
|
|
|
251 |
- If a question has multiple valid interpretations, choose the **narrowest, most literal** one.
|
252 |
- If the answer is not found, say `[ANSWER] - unknown`.
|
253 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
254 |
---
|
255 |
|
256 |
You must follow the examples (These answers are correct in case you see the similar questions):
|
|
|
287 |
|
288 |
return final_str
|
289 |
|
290 |
+
def evaluate_random_questions(self, csv_path: str = "gaia_extracted.csv", sample_size: int = 3, show_steps: bool = True):
|
291 |
+
import pandas as pd
|
292 |
+
from rich.table import Table
|
293 |
+
from rich.console import Console
|
294 |
+
|
295 |
df = pd.read_csv(csv_path)
|
296 |
if not {"question", "answer"}.issubset(df.columns):
|
297 |
print("CSV must contain 'question' and 'answer' columns.")
|
298 |
print("Found columns:", df.columns.tolist())
|
299 |
return
|
300 |
+
|
301 |
samples = df.sample(n=sample_size)
|
302 |
+
records = []
|
303 |
+
correct_count = 0
|
304 |
+
|
305 |
for _, row in samples.iterrows():
|
306 |
+
taskid = row["taskid"].strip()
|
307 |
question = row["question"].strip()
|
308 |
+
expected = str(row['answer']).strip()
|
309 |
+
agent_answer = self("taskid: " + taskid + ",\nquestion: " + question).strip()
|
310 |
+
|
311 |
+
is_correct = (expected == agent_answer)
|
312 |
+
correct_count += is_correct
|
313 |
+
records.append((question, expected, agent_answer, "✓" if is_correct else "✗"))
|
314 |
+
|
315 |
if show_steps:
|
316 |
print("---")
|
317 |
print("Question:", question)
|
318 |
print("Expected:", expected)
|
319 |
+
print("Agent:", agent_answer)
|
320 |
+
print("Correct:", is_correct)
|
321 |
+
|
322 |
+
# Print result table
|
323 |
+
console = Console()
|
324 |
+
table = Table(show_lines=True)
|
325 |
+
table.add_column("Question", overflow="fold")
|
326 |
+
table.add_column("Expected")
|
327 |
+
table.add_column("Agent")
|
328 |
+
table.add_column("Correct")
|
329 |
+
|
330 |
+
for question, expected, agent_ans, correct in records:
|
331 |
+
table.add_row(question, expected, agent_ans, correct)
|
332 |
+
|
333 |
+
console.print(table)
|
334 |
+
percent = (correct_count / sample_size) * 100
|
335 |
+
print(f"\nTotal Correct: {correct_count} / {sample_size} ({percent:.2f}%)")
|
336 |
+
|
337 |
|
338 |
if __name__ == "__main__":
|
339 |
args = sys.argv[1:]
|