Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -4,8 +4,7 @@ import gradio as gr
|
|
4 |
import requests
|
5 |
import pandas as pd
|
6 |
|
7 |
-
from smolagents import
|
8 |
-
|
9 |
from audio_transcriber import AudioTranscriptionTool
|
10 |
from image_analyzer import ImageAnalysisTool
|
11 |
from wikipedia_searcher import WikipediaSearcher
|
@@ -13,46 +12,68 @@ from wikipedia_searcher import WikipediaSearcher
|
|
13 |
# GAIA scoring endpoint
|
14 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
15 |
|
16 |
-
#
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
Rules to follow:
|
19 |
1. Return only the exact requested answer: no explanation and no reasoning.
|
20 |
2. For yes/no questions, return exactly \"Yes\" or \"No\".
|
21 |
3. For dates, use the exact format requested.
|
22 |
4. For numbers, use the exact number, no other format.
|
23 |
5. For names, use the exact name as found in sources.
|
24 |
-
6. If the question has an associated file,
|
25 |
Examples of good responses:
|
26 |
- \"42\"
|
|
|
27 |
- \"Yes\"
|
28 |
- \"October 5, 2001\"
|
29 |
- \"Buenos Aires\"
|
30 |
Never include phrases like \"the answer is...\" or \"Based on my research\".
|
31 |
-
Only return the exact answer.
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
image_tool = ImageAnalysisTool()
|
36 |
-
wikipedia_tool = WikipediaSearcher()
|
37 |
-
|
38 |
-
tools = [audio_tool, image_tool, wikipedia_tool]
|
39 |
-
|
40 |
-
# Define the custom agent using Dolphin model (free Mixtral)
|
41 |
-
class MyAgent(CodeAgent):
|
42 |
-
def __init__(self):
|
43 |
-
model = HfApiModel(
|
44 |
-
model="cognitivecomputations/dolphin-2.6-mixtral-8x7b",
|
45 |
-
api_key=os.getenv("HF_API_TOKEN", "").strip(),
|
46 |
-
# No system_prompt here
|
47 |
-
)
|
48 |
-
super().__init__(model=model, tools=tools)
|
49 |
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
|
57 |
# Evaluation + Submission function
|
58 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
@@ -70,7 +91,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
70 |
submit_url = f"{api_url}/submit"
|
71 |
|
72 |
try:
|
73 |
-
agent =
|
74 |
except Exception as e:
|
75 |
print(f"Error initializing agent: {e}")
|
76 |
return f"Error initializing agent: {e}", None
|
@@ -97,7 +118,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
97 |
if not task_id:
|
98 |
continue
|
99 |
try:
|
100 |
-
submitted_answer = agent(item)
|
101 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
102 |
results_log.append({
|
103 |
"Task ID": task_id,
|
|
|
4 |
import requests
|
5 |
import pandas as pd
|
6 |
|
7 |
+
from smolagents import InferenceClientModel, ToolCallingAgent
|
|
|
8 |
from audio_transcriber import AudioTranscriptionTool
|
9 |
from image_analyzer import ImageAnalysisTool
|
10 |
from wikipedia_searcher import WikipediaSearcher
|
|
|
12 |
# GAIA scoring endpoint
|
13 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
14 |
|
15 |
+
# Define the GaiaAgent class with embedded prompt in __call__
|
16 |
+
class GaiaAgent:
|
17 |
+
def __init__(self):
|
18 |
+
print("Gaia Agent Initialized")
|
19 |
+
|
20 |
+
self.model = InferenceClientModel(
|
21 |
+
model_id="cognitivecomputations/dolphin-2.6-mixtral-8x7b",
|
22 |
+
token=os.getenv("HUGGINGFACEHUB_API_TOKEN", "").strip()
|
23 |
+
)
|
24 |
+
|
25 |
+
self.tools = [
|
26 |
+
AudioTranscriptionTool(),
|
27 |
+
ImageAnalysisTool(),
|
28 |
+
WikipediaSearcher()
|
29 |
+
]
|
30 |
+
|
31 |
+
self.agent = ToolCallingAgent(
|
32 |
+
tools=self.tools,
|
33 |
+
model=self.model
|
34 |
+
)
|
35 |
+
|
36 |
+
def __call__(self, question: str) -> str:
|
37 |
+
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
38 |
+
|
39 |
+
prompt = f"""You are an agent solving the GAIA benchmark and you are required to provide exact answers.
|
40 |
Rules to follow:
|
41 |
1. Return only the exact requested answer: no explanation and no reasoning.
|
42 |
2. For yes/no questions, return exactly \"Yes\" or \"No\".
|
43 |
3. For dates, use the exact format requested.
|
44 |
4. For numbers, use the exact number, no other format.
|
45 |
5. For names, use the exact name as found in sources.
|
46 |
+
6. If the question has an associated file, download the file first using the task ID.
|
47 |
Examples of good responses:
|
48 |
- \"42\"
|
49 |
+
- \"Arturo Nunez\"
|
50 |
- \"Yes\"
|
51 |
- \"October 5, 2001\"
|
52 |
- \"Buenos Aires\"
|
53 |
Never include phrases like \"the answer is...\" or \"Based on my research\".
|
54 |
+
Only return the exact answer.
|
55 |
+
QUESTION:
|
56 |
+
{question}
|
57 |
+
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
|
59 |
+
try:
|
60 |
+
result = self.agent.run(prompt)
|
61 |
+
print(f"Raw result from agent: {result}")
|
62 |
+
|
63 |
+
if isinstance(result, dict) and "answer" in result:
|
64 |
+
return str(result["answer"]).strip()
|
65 |
+
elif isinstance(result, str):
|
66 |
+
return result.strip()
|
67 |
+
elif isinstance(result, list):
|
68 |
+
for item in reversed(result):
|
69 |
+
if isinstance(item, dict) and item.get("role") == "assistant" and "content" in item:
|
70 |
+
return item["content"].strip()
|
71 |
+
return "ERROR: Unexpected list format"
|
72 |
+
else:
|
73 |
+
return "ERROR: Unexpected result type"
|
74 |
+
except Exception as e:
|
75 |
+
print(f"Exception during agent run: {e}")
|
76 |
+
return f"AGENT ERROR: {e}"
|
77 |
|
78 |
# Evaluation + Submission function
|
79 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
|
91 |
submit_url = f"{api_url}/submit"
|
92 |
|
93 |
try:
|
94 |
+
agent = GaiaAgent()
|
95 |
except Exception as e:
|
96 |
print(f"Error initializing agent: {e}")
|
97 |
return f"Error initializing agent: {e}", None
|
|
|
118 |
if not task_id:
|
119 |
continue
|
120 |
try:
|
121 |
+
submitted_answer = agent(item.get("question", ""))
|
122 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
123 |
results_log.append({
|
124 |
"Task ID": task_id,
|