File size: 14,497 Bytes
10e9b7d
 
eccf8e4
3c4371f
8aedee6
 
 
b6137ed
8aedee6
 
 
 
b6137ed
10e9b7d
e80aab9
3db6293
b6137ed
 
e80aab9
8aedee6
 
 
 
 
 
 
 
 
 
 
 
 
2ae28e5
 
 
 
 
b6137ed
 
 
8aedee6
 
 
 
 
31243f4
8aedee6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6137ed
 
8aedee6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6137ed
 
 
8aedee6
 
 
 
 
 
 
b6137ed
8aedee6
 
b6137ed
8aedee6
 
b6137ed
8aedee6
 
2ae28e5
8aedee6
 
 
 
 
2ae28e5
 
 
8aedee6
2ae28e5
8aedee6
 
 
 
 
 
2ae28e5
b6137ed
 
 
 
 
 
 
 
 
 
 
 
 
 
2ae28e5
 
 
 
 
 
b6137ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8aedee6
 
 
 
 
 
 
 
 
 
 
 
 
b6137ed
 
8aedee6
 
 
b6137ed
8aedee6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6137ed
8aedee6
 
 
 
 
 
b6137ed
8aedee6
b6137ed
8aedee6
 
b6137ed
8aedee6
 
 
 
 
 
b6137ed
8aedee6
 
 
b6137ed
8aedee6
 
 
 
 
 
 
 
 
b6137ed
8aedee6
 
 
 
 
 
 
b6137ed
8aedee6
49bfb00
8aedee6
b6137ed
8aedee6
 
 
 
 
7e4a06b
b6137ed
e80aab9
31243f4
8aedee6
31243f4
 
8aedee6
eccf8e4
b6137ed
7d65c66
31243f4
7d65c66
8aedee6
e80aab9
b6137ed
31243f4
b6137ed
 
31243f4
b6137ed
 
 
31243f4
7d65c66
31243f4
 
b6137ed
31243f4
7d65c66
8aedee6
e80aab9
b6137ed
e80aab9
 
31243f4
b6137ed
 
e80aab9
b6137ed
e80aab9
b6137ed
 
e80aab9
8aedee6
e80aab9
8aedee6
0ee0419
e514fd7
 
b6137ed
 
 
e514fd7
e80aab9
8aedee6
 
 
b6137ed
7d65c66
b6137ed
e80aab9
 
8aedee6
b6137ed
8aedee6
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
import os
import gradio as gr
import requests
import pandas as pd
import re
import io
import contextlib
import json
from huggingface_hub import InferenceClient
from langchain_community.tools import DuckDuckGoSearchRun
from PyPDF2 import PdfReader
from docx import Document
from youtube_transcript_api import YouTubeTranscriptApi

# --- Constants ---
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
MODEL_ID = "mistralai/Mistral-7B-Instruct-v0.2"
PROMPT_TEMPLATE = """<s>[INST]You are a helpful assistant designed to answer questions accurately. You have access to the following tools:

{tools_description}

To answer the question, you must follow this format, thinking step by step.

Thought: Your reasoning and plan for the next step. You can also write down observations here.
Action: The tool to use, in the format `tool_name(arg_name="value")`. The available tools are: {tool_names}.
Observation: The result from the tool.
... (this Thought/Action/Observation can repeat N times)

When you have the final answer, respond with:
Thought: I have now found the final answer.
Final Answer: The final answer.

Important:
- Do not use a tool if you are not sure about the parameters.
- Do not make up file names.
- If a tool is not available for a task (e.g., image analysis), state that you cannot answer.
- If a tool returns an error, note it and try an alternative approach if possible.

Question: {question}
[/INST]{scratchpad}"""

# --- Tool Definitions ---

class WebSearchTool:
    """A tool to search the web for information."""
    def __init__(self):
        self.search = DuckDuckGoSearchRun()

    def __call__(self, query: str):
        print(f"--- Calling WebSearchTool with query: {query} ---")
        try:
            return self.search.run(query)
        except Exception as e:
            return f"Error during web search: {e}"

    @property
    def description(self):
        return 'web_search(query: str) -> str - A tool to search the web for information. Use it to find up-to-date information or facts.'

class PythonREPLTool:
    """A tool to execute Python code."""
    def __call__(self, code: str):
        print(f"--- Calling PythonREPLTool with code: {code} ---")
        if any(keyword in code for keyword in ["os", "sys", "subprocess", "eval", "exec"]):
            return "Error: Use of os, sys, subprocess, eval, or exec is not allowed for security reasons."
            
        local_vars = {}
        string_io = io.StringIO()
        try:
            with contextlib.redirect_stdout(string_io):
                exec(code, {}, local_vars)
            output = string_io.getvalue()
            if not output and local_vars:
                output = str(list(local_vars.values())[-1])
            return output if output else "Code executed with no output."
        except Exception as e:
            return f"Error executing code: {e}"

    @property
    def description(self):
        return 'python_repl(code: str) -> str - A Python REPL. Use it to perform calculations, data manipulation, etc. The result of the last line is returned.'

class FileReaderTool:
    """A tool to read the content of a file associated with a task."""
    def __init__(self, api_url: str):
        self.api_url = api_url

    def __call__(self, task_id: str, file_name: str):
        print(f"--- Calling FileReaderTool for task_id: {task_id}, file_name: {file_name} ---")
        if file_name.endswith(('.mp3', '.wav', '.flac')):
            return "Error: This tool cannot read audio files. Use the `audio_transcriber` tool instead."
            
        file_url = f"{self.api_url}/files/{task_id}"
        
        try:
            response = requests.get(file_url, timeout=20)
            response.raise_for_status()
            file_content = io.BytesIO(response.content)
            
            content = ""
            if file_name.endswith('.pdf'):
                pdf = PdfReader(file_content)
                content = "".join(page.extract_text() for page in pdf.pages if page.extract_text())
            elif file_name.endswith('.docx'):
                doc = Document(file_content)
                content = "\n".join(para.text for para in doc.paragraphs)
            elif file_name.endswith('.csv'):
                df = pd.read_csv(file_content)
                content = df.to_json(orient='records')  # Return JSON for easier processing
            elif file_name.endswith('.json'):
                data = json.load(file_content)
                content = json.dumps(data, indent=2)
            elif file_name.endswith('.txt'):
                content = file_content.read().decode('utf-8')
            elif file_name.endswith('.xlsx'):
                df = pd.read_excel(file_content, engine='openpyxl')
                content = df.to_json(orient='records')  # Return JSON for easier processing
            else:
                return f"Error: Unsupported file type for '{file_name}'. Supported types: .pdf, .docx, .csv, .json, .txt, .xlsx."
            return content if content else "File is empty."
        except Exception as e:
            return f"Error reading file '{file_name}': {e}"

    @property
    def description(self):
        return 'file_reader(task_id: str, file_name: str) -> str - Reads content of text-based files (.pdf, .docx, .csv, .json, .txt, .xlsx). For audio, use audio_transcriber.'

class AudioTranscriptionTool:
    """A tool to transcribe audio files using the Hugging Face Inference API."""
    def __init__(self, api_url: str, client: InferenceClient):
        self.api_url = api_url
        self.client = client

    def __call__(self, task_id: str, file_name: str):
        print(f"--- Calling AudioTranscriptionTool for task: {task_id}, file: {file_name} ---")
        file_url = f"{self.api_url}/files/{task_id}"
        try:
            response = requests.get(file_url, timeout=30)
            response.raise_for_status()
            audio_data = response.content
            # Specify Whisper-large-v2 for accurate transcription
            transcription = self.client.automatic_speech_recognition(audio_data, model="openai/whisper-large-v2")
            if transcription and 'text' in transcription:
                return transcription['text']
            else:
                return "Could not transcribe audio."
        except Exception as e:
            return f"Error during audio transcription: {e}"

    @property
    def description(self):
        return 'audio_transcriber(task_id: str, file_name: str) -> str - Transcribes an audio file (.mp3, .wav) associated with the current task.'

class YouTubeTranscriptTool:
    """A tool to fetch the transcript of a YouTube video."""
    def __call__(self, video_url: str):
        print(f"--- Calling YouTubeTranscriptTool for URL: {video_url} ---")
        match = re.search(r"v=([a-zA-Z0-9_-]+)", video_url)
        if not match:
            return "Error: Invalid YouTube URL. Could not extract video ID."
        video_id = match.group(1)
        try:
            transcript_list = YouTubeTranscriptApi.get_transcript(video_id)
            return " ".join([d['text'] for d in transcript_list])
        except Exception as e:
            return f"Error fetching transcript for video {video_id}: {e}. The video might not have a transcript."

    @property
    def description(self):
        return 'youtube_transcript_fetcher(video_url: str) -> str - Fetches the transcript of a YouTube video. Use for questions about video content.'

# --- GAIA Agent Definition ---
class GaiaAgent:
    def __init__(self, hf_token: str, api_url: str, max_turns: int = 8):
        if not hf_token:
            raise ValueError("Hugging Face token is required for the Inference API.")
        self.llm_client = InferenceClient(model=MODEL_ID, token=hf_token)
        self.max_turns = max_turns
        
        self.tools = {
            "web_search": WebSearchTool(),
            "python_repl": PythonREPLTool(),
            "file_reader": FileReaderTool(api_url=api_url),
            "youtube_transcript_fetcher": YouTubeTranscriptTool(),
            "audio_transcriber": AudioTranscriptionTool(api_url=api_url, client=self.llm_client),
        }
        self.tools_description = "\n".join([f"- `{tool.description}`" for tool in self.tools.values()])
        self.tool_names = ", ".join(self.tools.keys())
        print("GaiaAgent initialized successfully with tools:", self.tool_names)

    def __call__(self, question: str, task_id: str) -> str:
        print(f"\n--- Running agent on task {task_id} ---")
        print(f"Question: {question[:100]}...")
        scratchpad = ""
        
        for turn in range(self.max_turns):
            print(f"Turn {turn + 1}/{self.max_turns}")
            prompt = PROMPT_TEMPLATE.format(
                tools_description=self.tools_description,
                tool_names=self.tool_names,
                question=question,
                scratchpad=scratchpad,
            )
            try:
                llm_output = self.llm_client.text_generation(
                    prompt, max_new_tokens=1024, stop_sequences=["Observation:", "[/INST]"], temperature=0.1
                ).strip()
            except Exception as e:
                return f"Error: LLM call failed. {e}"

            print(f"LLM Output:\n{llm_output}")
            scratchpad += llm_output
            
            final_answer_match = re.search(r"Final Answer:\s*(.*)", scratchpad, re.DOTALL)
            action_match = re.search(r"Action:\s*([a-zA-Z0-9_]+)\((.*)\)", llm_output, re.DOTALL)

            if final_answer_match:
                return final_answer_match.group(1).strip()
            
            elif action_match:
                tool_name = action_match.group(1).strip()
                tool_args_str = action_match.group(2).strip()
                
                if tool_name not in self.tools:
                    observation = f"Error: Unknown tool '{tool_name}'."
                else:
                    try:
                        args_dict = eval(f"dict({tool_args_str})", {"__builtins__": None}, {})
                        if tool_name in ['file_reader', 'audio_transcriber']:
                            args_dict['task_id'] = task_id
                        tool = self.tools[tool_name]
                        observation = tool(**args_dict)
                    except Exception as e:
                        observation = f"Error executing tool '{tool_name}': {e}"
                
                print(f"Observation: {str(observation)[:200]}...")
                scratchpad += f"\nObservation: {str(observation)}\n"
            else:
                scratchpad += "\nObservation: No valid action or final answer found. Please format your response as either 'Action: tool_name(args)' or 'Final Answer: your_answer'."

        return "Agent stopped after reaching maximum turns."

# --- Main Submission Logic ---
def run_and_submit_all(profile: gr.OAuthProfile | None):
    hf_token = os.getenv("HF_TOKEN")
    if not hf_token:
        return "Error: `HF_TOKEN` secret not set. Please add it to your Space secrets.", None

    space_id = "ZeroTimo/RobotPai"
    if not space_id:
        return "Error: `SPACE_ID` not found. Are you in a Hugging Face Space?", None

    if not profile:
        return "Please Login to Hugging Face with the button to submit.", None
    
    username = profile.username
    api_url = DEFAULT_API_URL
    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"

    try:
        agent = GaiaAgent(hf_token=hf_token, api_url=api_url)
    except Exception as e:
        return f"Error initializing agent: {e}", None

    try:
        response = requests.get(f"{api_url}/questions", timeout=15)
        response.raise_for_status()
        questions_data = response.json()
    except Exception as e:
        return f"Error fetching questions: {e}", None

    results_log, answers_payload = [], []
    for item in questions_data:
        task_id, question_text = item.get("task_id"), item.get("question")
        if not all([task_id, question_text]): continue
        try:
            answer = agent(question_text, task_id)
            answers_payload.append({"task_id": task_id, "submitted_answer": answer})
            results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": answer})
        except Exception as e:
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})

    if not answers_payload:
        return "Agent did not produce any answers.", pd.DataFrame(results_log)

    submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
    
    try:
        response = requests.post(f"{api_url}/submit", json=submission_data, timeout=120)
        response.raise_for_status()
        result_data = response.json()
        final_status = (
            f"Submission Successful! Score: {result_data.get('score', 'N/A')}% "
            f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')})"
        )
        return final_status, pd.DataFrame(results_log)
    except requests.exceptions.RequestException as e:
        error_detail = f"Server responded with status {e.response.status_code}. Response: {e.response.text[:500]}" if e.response else str(e)
        return f"Submission Failed: {error_detail}", pd.DataFrame(results_log)

# --- Gradio Interface ---
with gr.Blocks() as demo:
    gr.Markdown("# GAIA Agent Evaluation Runner")
    gr.Markdown(
        """
        **Instructions:**
        1.  **Add your HF Token**: Go to your Space's **Settings** and add a secret named `HF_TOKEN` with your Hugging Face `read` token.
        2.  **Login**: Use the button below to login with your Hugging Face account.
        3.  **Run**: Click 'Run Evaluation & Submit' to start the agent. This may take several minutes.
        """
    )
    with gr.Row():
        gr.LoginButton()
        run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
    status_output = gr.Textbox(label="Run Status / Submission Result", lines=4, interactive=False)
    results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
    run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])

if __name__ == "__main__":
    if not os.getenv("HF_TOKEN"):
        print("⚠️ WARNING: `HF_TOKEN` secret not found. The agent will not run.")
    demo.launch()