File size: 17,841 Bytes
ca2b63a
 
 
26e4907
574b6ca
 
 
 
a42d6f7
51e7f46
26e4907
 
 
 
10e9b7d
a42d6f7
 
 
 
 
 
 
 
 
 
 
 
 
 
757ebd9
e80aab9
3db6293
26e4907
e80aab9
ca2b63a
 
31243f4
26e4907
 
a42d6f7
26e4907
51e7f46
26e4907
 
 
51e7f46
 
26e4907
 
 
 
 
51e7f46
26e4907
 
a42d6f7
26e4907
ca2b63a
 
26e4907
ca2b63a
26e4907
ca2b63a
 
26e4907
ca2b63a
26e4907
ca2b63a
 
a42d6f7
26e4907
51e7f46
 
 
 
 
26e4907
 
 
 
 
51e7f46
26e4907
51e7f46
26e4907
51e7f46
ca2b63a
26e4907
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a42d6f7
 
26e4907
a42d6f7
757ebd9
 
26e4907
 
 
 
 
 
 
 
 
 
 
 
c549c70
26e4907
 
 
 
 
 
 
 
 
 
 
757ebd9
26e4907
ca2b63a
26e4907
 
 
 
a42d6f7
26e4907
 
 
 
 
a42d6f7
26e4907
 
 
 
 
 
 
 
 
 
 
 
ca2b63a
26e4907
 
 
c549c70
26e4907
 
c549c70
26e4907
 
 
 
 
 
 
c549c70
26e4907
 
 
 
c549c70
26e4907
 
 
 
c549c70
ca2b63a
26e4907
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c549c70
26e4907
51e7f46
ca2b63a
26e4907
 
 
 
 
 
 
 
 
 
 
 
 
c549c70
26e4907
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c549c70
26e4907
 
c549c70
26e4907
 
 
 
 
c549c70
26e4907
 
 
 
 
 
 
 
c549c70
26e4907
 
 
 
 
 
 
 
 
4021bf3
26e4907
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
757ebd9
26e4907
 
 
 
 
51e7f46
 
a42d6f7
ca2b63a
26e4907
757ebd9
3c4371f
7e4a06b
757ebd9
3c4371f
7e4a06b
3c4371f
7d65c66
3c4371f
7e4a06b
31243f4
 
e80aab9
26e4907
31243f4
757ebd9
31243f4
26e4907
31243f4
757ebd9
36ed51a
a42d6f7
3c4371f
757ebd9
31243f4
eccf8e4
31243f4
7d65c66
31243f4
 
26e4907
31243f4
7d65c66
26e4907
e80aab9
26e4907
7d65c66
 
a42d6f7
 
31243f4
26e4907
a42d6f7
26e4907
31243f4
a42d6f7
26e4907
a42d6f7
31243f4
26e4907
 
 
 
 
a42d6f7
26e4907
 
 
a42d6f7
51e7f46
26e4907
 
 
51e7f46
31243f4
26e4907
 
 
 
 
a42d6f7
26e4907
 
 
a42d6f7
31243f4
26e4907
a42d6f7
26e4907
 
a42d6f7
 
 
26e4907
e80aab9
7d65c66
e80aab9
26e4907
a42d6f7
26e4907
 
 
 
 
 
e80aab9
26e4907
a42d6f7
7d65c66
26e4907
 
 
e80aab9
757ebd9
 
a42d6f7
26e4907
 
 
 
 
a42d6f7
 
 
26e4907
 
 
 
a42d6f7
26e4907
 
 
a42d6f7
 
 
26e4907
 
 
a42d6f7
e80aab9
26e4907
31243f4
26e4907
e80aab9
 
757ebd9
e80aab9
a42d6f7
26e4907
 
a42d6f7
 
 
 
26e4907
a42d6f7
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
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
from llama_index.llms.huggingface import HuggingFaceLLM
from llama_index.core.agent import ReActAgent
from llama_index.core.tools import FunctionTool
from transformers import AutoTokenizer, pipeline
import os
import gradio as gr
import requests
import pandas as pd
import traceback
import torch
import re
import gc
from typing import List, Dict
from datetime import datetime

# Import real tool dependencies
try:
    from duckduckgo_search import DDGS
except ImportError:
    print("Warning: duckduckgo_search not installed. Web search will be limited.")
    DDGS = None

try:
    from sympy import sympify
    from sympy.core.sympify import SympifyError
except ImportError:
    print("Warning: sympy not installed. Math calculator will be limited.")
    sympify = None
    SympifyError = Exception

# --- Constants ---
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
MEMORY_LIMIT_GB = 16  # Your system's memory limit

# --- Advanced Agent Definition ---
class SmartAgent:
    def __init__(self):
        print(f"Initializing Local LLM Agent (Memory Limit: {MEMORY_LIMIT_GB}GB)...")
        self.model_loaded = False
        
        # Model options sorted by capability (name, approx size in GB, quantization)
        model_options = [
            ("google/flan-t5-large", 3, "8-bit"),  # Best balance for 16GB
            ("google/flan-t5-base", 1, "8-bit"),    # Smaller fallback
            ("facebook/opt-1.3b", 2.5, "8-bit")     # Alternative option
        ]
        
        # Try loading models until success
        for model_name, size_gb, quantization in model_options:
            if size_gb <= MEMORY_LIMIT_GB and self._try_load_model(model_name, quantization):
                self.model_loaded = True
                break
        
        if not self.model_loaded:
            raise RuntimeError("Failed to load any suitable model within memory constraints")
        
        # Initialize tools with enhanced implementations
        self.tools = [
            FunctionTool.from_defaults(
                fn=self.smart_web_search,
                name="web_search",
                description="Searches the web for current information. Use for questions about recent events, people, or facts not in the model's training data."
            ),
            FunctionTool.from_defaults(
                fn=self.robust_math_calculator,
                name="math_calculator",
                description="Solves mathematical expressions and equations. Use for calculations, arithmetic, algebra, or numerical problems."
            )
        ]
        
        # Initialize ReAct agent with memory optimization
        try:
            self.agent = ReActAgent.from_tools(
                tools=self.tools,
                llm=self.llm,
                verbose=True,
                max_iterations=4,
                react_context="""Think step by step. Use tools when needed:
                - For current/recent information: web_search
                - For calculations: math_calculator
                - Be concise but accurate"""
            )
            print("ReAct Agent initialized successfully")
        except Exception as e:
            print(f"ReAct Agent init failed: {e}")
            self.agent = None

    def _try_load_model(self, model_name: str, quantization: str) -> bool:
        """Attempt to load model with memory constraints"""
        try:
            print(f"Loading {model_name} with {quantization} quantization...")
            
            model_kwargs = {
                "torch_dtype": torch.float16,
                "low_cpu_mem_usage": True,
            }
            
            if quantization == "8-bit":
                model_kwargs["load_in_8bit"] = True
            elif quantization == "4-bit":
                model_kwargs["load_in_4bit"] = True
            
            self.llm = HuggingFaceLLM(
                model_name=model_name,
                tokenizer_name=model_name,
                context_window=2048,
                max_new_tokens=256,
                generate_kwargs={
                    "temperature": 0.4,
                    "do_sample": True,
                    "top_p": 0.9,
                    "repetition_penalty": 1.1
                },
                device_map="auto" if torch.cuda.is_available() else "cpu",
                model_kwargs=model_kwargs
            )
            
            # Test the model
            test_response = self.llm.complete("Test response:")
            if not test_response:
                raise ValueError("Model failed test response")
                
            print(f"Successfully loaded {model_name}")
            return True
            
        except Exception as e:
            print(f"Failed to load {model_name}: {str(e)}")
            self.cleanup_memory()
            return False

    def smart_web_search(self, query: str) -> str:
        """Enhanced web search with focused results"""
        print(f"Searching: {query[:60]}...")
        
        if not DDGS:
            return "Web search unavailable (duckduckgo_search not installed)"
        
        try:
            with DDGS() as ddgs:
                # Get focused results with longer snippets
                results = list(ddgs.text(query, max_results=3))
                
                if not results:
                    return "No results found"
                
                # Process results for key information
                processed = []
                for i, res in enumerate(results, 1):
                    title = res.get('title', 'No title')
                    body = res.get('body', 'No description')
                    url = res.get('href', '')
                    
                    # Extract most relevant part for the query
                    key_info = self._extract_relevant_info(query, body)
                    
                    processed.append(
                        f"πŸ” Result {i}:\n"
                        f"Title: {title}\n"
                        f"Info: {key_info[:250]}\n"
                        f"Source: {url}\n"
                    )
                
                return "\n".join(processed)
        except Exception as e:
            return f"Search error: {str(e)}"

    def _extract_relevant_info(self, query: str, text: str) -> str:
        """Extract the most relevant portion of text for the query"""
        query_lower = query.lower()
        text_lower = text.lower()
        
        # Handle different question types
        if any(w in query_lower for w in ['who is', 'biography', 'born']):
            # Look for birth/death info
            match = re.search(r"(born [^.]+? in [^.]+?\.)", text, re.I)
            return match.group(1) if match else text[:250]
        
        elif any(w in query_lower for w in ['died', 'death']):
            match = re.search(r"(died [^.]+?\.)", text, re.I)
            return match.group(1) if match else text[:250]
        
        elif any(w in query_lower for w in ['award', 'prize', 'won']):
            match = re.search(r"(awarded [^.]+? in [^.]+?\.)", text, re.I)
            return match.group(1) if match else text[:250]
        
        # Default: return first 250 chars with important sentences
        sentences = re.split(r'(?<=[.!?]) +', text)
        important = [s for s in sentences if any(w in s.lower() for w in query.lower().split())]
        return " ".join(important[:3]) if important else text[:250]

    def robust_math_calculator(self, expression: str) -> str:
        """Improved math calculator with better parsing"""
        print(f"Calculating: {expression}")
        
        # Clean and preprocess the expression
        expr = expression.strip("'\"")
        
        # Replace words with operators
        replacements = {
            'plus': '+', 'minus': '-', 'times': '*', 'divided by': '/',
            '^': '**', 'percent': '/100', 'modulo': '%'
        }
        for word, op in replacements.items():
            expr = expr.replace(word, op)
        
        # Extract math expression from text
        math_match = re.search(r"([-+]?\d*\.?\d+[+\-*/%^()\s]+\d+\.?\d*)", expr)
        if math_match:
            expr = math_match.group(1)
        
        # Safety check
        allowed_chars = set("0123456789+-*/().%^ ")
        if not all(c in allowed_chars for c in expr.replace(" ", "")):
            return "Error: Invalid characters in expression"
        
        try:
            # Try direct evaluation first
            result = eval(expr)
            return f"Result: {result}"
        except:
            # Fallback to sympy if available
            if sympify:
                try:
                    result = sympify(expr).evalf()
                    return f"Result: {result}"
                except SympifyError as e:
                    return f"Math error: {str(e)}"
            return "Error: Could not evaluate the expression"

    def __call__(self, question: str) -> str:
        """Main interface for answering questions"""
        print(f"\nQuestion: {question[:100]}...")
        
        try:
            # Step 1: Classify question type
            q_type = self._classify_question(question)
            
            # Step 2: Use appropriate strategy
            if q_type == "fact":
                return self._answer_fact_question(question)
            elif q_type == "math":
                return self._answer_math_question(question)
            else:
                return self._answer_general_question(question)
                
        except Exception as e:
            print(f"Error processing question: {str(e)}")
            return self._fallback_response(question)

    def _classify_question(self, question: str) -> str:
        """Determine the type of question"""
        q_lower = question.lower()
        
        # Math questions
        math_keywords = ['calculate', 'compute', 'sum', 'total', 'average', 
                        'percentage', 'equation', 'solve', 'math', 'number',
                        '+', '-', '*', '/', '=']
        if any(kw in q_lower for kw in math_keywords):
            return "math"
        
        # Fact-based questions
        fact_keywords = ['current', 'latest', 'recent', 'today', 'news', 
                        'who is', 'what is', 'when did', 'where is',
                        'competition', 'winner', 'recipient', 'nationality',
                        'country', 'malko', 'century', 'award', 'born', 'died']
        if any(kw in q_lower for kw in fact_keywords):
            return "fact"
        
        return "general"

    def _answer_fact_question(self, question: str) -> str:
        """Handle fact-based questions with web search"""
        # Extract key entities for focused search
        entities = re.findall(r"([A-Z][a-z]+(?:\s+[A-Z][a-z]+)*)", question)
        search_query = " ".join(entities[:3]) or question[:50]
        
        # Get search results
        search_results = self.smart_web_search(search_query)
        
        # Process with LLM if available
        if self.model_loaded:
            prompt = f"""Question: {question}
            Search Results:
            {search_results}
            
            Based ONLY on these results, provide a concise answer.
            If the answer isn't there, say so."""
            
            try:
                response = self.llm.complete(prompt)
                return str(response).strip()
            except:
                return f"Search results for '{search_query}':\n{search_results}"
        
        return f"Search results for '{search_query}':\n{search_results}"

    def _answer_math_question(self, question: str) -> str:
        """Handle math questions with calculator"""
        # Try to extract math expression
        math_expr = re.search(r"([\d\s+\-*/().^]+)", question)
        if math_expr:
            return self.robust_math_calculator(math_expr.group(1))
        
        # If no clear expression, use agent reasoning
        if self.agent:
            try:
                response = self.agent.query(question)
                return str(response)
            except:
                return self._fallback_response(question)
        
        return self._fallback_response(question)

    def _answer_general_question(self, question: str) -> str:
        """Handle general knowledge questions"""
        if self.agent:
            try:
                response = self.agent.query(question)
                return str(response)
            except:
                return self._fallback_response(question)
        
        # Fallback to simple LLM response
        try:
            response = self.llm.complete(question)
            return str(response)
        except:
            return self._fallback_response(question)

    def _fallback_response(self, question: str) -> str:
        """Final fallback when all else fails"""
        return f"I couldn't generate a complete answer for: {question[:150]}... Please try rephrasing or ask about something more specific."

    def cleanup_memory(self):
        """Clean up memory resources"""
        if torch.cuda.is_available():
            torch.cuda.empty_cache()
        gc.collect()


# --- Submission Logic ---
def run_and_submit_all(profile: gr.OAuthProfile | None):
    """Handle the full evaluation process"""
    space_id = os.getenv("SPACE_ID")

    if profile:
        username = f"{profile.username}"
        print(f"User logged in: {username}")
    else:
        print("User not logged in.")
        return "Please Login to Hugging Face with the button.", None

    api_url = DEFAULT_API_URL
    questions_url = f"{api_url}/questions"
    submit_url = f"{api_url}/submit"

    # Initialize agent with memory management
    try:
        agent = SmartAgent()
    except Exception as e:
        print(f"Agent initialization failed: {e}")
        return f"Error initializing agent: {e}", None

    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
    print(f"Agent code URL: {agent_code}")

    # Fetch Questions
    print(f"Fetching questions from: {questions_url}")
    try:
        response = requests.get(questions_url, timeout=15)
        response.raise_for_status()
        questions_data = response.json()
        if not questions_data:
            return "No questions received from server.", None
        print(f"Fetched {len(questions_data)} questions.")
    except Exception as e:
        return f"Error fetching questions: {e}", None

    # Process Questions
    results_log = []
    answers_payload = []
    
    for i, item in enumerate(questions_data, 1):
        task_id = item.get("task_id")
        question = item.get("question")
        
        if not task_id or not question:
            continue
            
        print(f"Processing question {i}/{len(questions_data)} (ID: {task_id})")
        
        try:
            answer = agent(question)
            answers_payload.append({
                "task_id": task_id,
                "submitted_answer": answer[:2000]  # Limit answer length
            })
            results_log.append({
                "Task ID": task_id,
                "Question": question[:100] + "..." if len(question) > 100 else question,
                "Answer": answer[:200] + "..." if len(answer) > 200 else answer
            })
            
            # Clean memory every 5 questions
            if i % 5 == 0:
                agent.cleanup_memory()
                
        except Exception as e:
            print(f"Error on question {task_id}: {e}")
            answers_payload.append({
                "task_id": task_id,
                "submitted_answer": f"Error processing question: {str(e)}"
            })
            results_log.append({
                "Task ID": task_id,
                "Question": question[:100] + "..." if len(question) > 100 else question,
                "Answer": f"Error: {str(e)}"
            })

    # Submit Answers
    submission_data = {
        "username": username.strip(),
        "agent_code": agent_code,
        "answers": answers_payload
    }
    
    print(f"Submitting {len(answers_payload)} answers...")
    try:
        response = requests.post(submit_url, json=submission_data, timeout=60)
        response.raise_for_status()
        result = response.json()
        
        status = (
            f"βœ… Submission Successful!\n\n"
            f"User: {result.get('username')}\n"
            f"Score: {result.get('score', 'N/A')}% "
            f"({result.get('correct_count', '?')}/{result.get('total_attempted', '?')})\n"
            f"Message: {result.get('message', '')}"
        )
        return status, pd.DataFrame(results_log)
        
    except Exception as e:
        error_msg = f"❌ Submission Failed: {str(e)}"
        print(error_msg)
        return error_msg, pd.DataFrame(results_log)


# --- Gradio UI ---
with gr.Blocks(title="Local LLM Agent Evaluation") as demo:
    gr.Markdown("""
    # οΏ½ Local LLM Agent Evaluation
    **Run your local agent against the course evaluation questions**
    """)
    
    with gr.Row():
        gr.LoginButton()
    
    run_btn = gr.Button(
        "πŸš€ Run Evaluation & Submit Answers",
        variant="primary"
    )
    
    status_out = gr.Textbox(
        label="πŸ“‹ Status",
        interactive=False
    )
    
    results_table = gr.DataFrame(
        label="πŸ“Š Results",
        interactive=False,
        wrap=True
    )

    run_btn.click(
        fn=run_and_submit_all,
        outputs=[status_out, results_table]
    )


if __name__ == "__main__":
    print("\n" + "="*60)
    print(f"πŸš€ Starting Agent Evaluation - {datetime.now().strftime('%Y-%m-%d %H:%M')}")
    print(f"Memory Limit: {MEMORY_LIMIT_GB}GB")
    print("="*60)
    
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860
    )