File size: 33,498 Bytes
01f199c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
from typing import List
import tiktoken
import os
import json
import re
import sys
import time

from copy import deepcopy
import xml.etree.ElementTree as ET

from .Base import BaseStrategy
from models.Base import BaseModel
from models.Pangu import Pangu

from datasets.Dataset import Dataset
from datasets.APPSDataset import APPSDataset
from datasets.MBPPDataset import MBPPDataset
from datasets.XCodeDataset import XCodeDataset
from datasets.HumanEvalDataset import HumanDataset
from datasets.CodeContestDataset import CodeContestDataset

from results.Results import Results
from evaluations.func_evaluate import evaluate_io

mapping = {
    1: "one (01)",
    2: "two (02)",
    3: "three (03)",
    4: "four (04)",
    5: "five (05)",
    6: "six (06)",
    7: "seven (07)",
    8: "eight (08)",
    9: "nine (09)",
}

# KB + Exemplars + Example Planning + Problem Planning + Code Generation + Sample IO testing + Code Improvement


class MapCoder(BaseStrategy):
    def __init__(
        self,
        k: int = 3,
        t: int = 5,
        pr_tok: int = 0,
        com_tok: int = 0,
        *args,
        **kwargs
    ):
        super().__init__(*args, **kwargs)
        self.k = k
        self.t = t
        self.pr_tok = 0
        self.com_tok = 0

    def xml_to_dict(self, element):
        result = {}
        for child in element:
            if child:
                child_data = self.xml_to_dict(child)
                if child.tag in result:
                    if isinstance(result[child.tag], list):
                        result[child.tag].append(child_data)
                    else:
                        result[child.tag] = [result[child.tag], child_data]
                else:
                    result[child.tag] = child_data
            else:
                result[child.tag] = child.text
        return result
    
    def remove_before_root(self, response: str) -> str:
        start_index = response.find('<root>')
        if start_index != -1:
            return response[start_index:]
        return response

    def parse_xml(self, response: str) -> dict:
        if '```xml' in response:
            response = response.replace('```xml', '')
        if '```' in response:
            response = response.replace('```', '')
        
        # 删除pangu返回时会出现的<root>前的多余内容
        response = self.remove_before_root(response)

        try:
            root = ET.fromstring(response)
        except:
            try:
                root = ET.fromstring('<root>\n' + response + '\n</root>')
            except:
                root = ET.fromstring('<root>\n' + response)
        mid = self.xml_to_dict(root)
        for k,v in mid.items():
            print(f"{k}")
        # sys.exit(0)
        return mid

    def parse_code(self, response: str) -> str:
        if "```" not in response:
            return response

        code_pattern = r'```((.|\n)*?)```'
        if "```Python" in response:
            code_pattern = r'```Python((.|\n)*?)```'
        if "```Python3" in response:
            code_pattern = r'```Python3((.|\n)*?)```'
        if "```python" in response:
            code_pattern = r'```python((.|\n)*?)```'
        if "```python3" in response:
            code_pattern = r'```python3((.|\n)*?)```'
        if "```C" in response:
            code_pattern = r'```C((.|\n)*?)```'
        if "```c" in response:
            code_pattern = r'```c((.|\n)*?)```'
        if "```C++" in response:
            code_pattern = r'```C\+\+((.|\n)*?)```'
        if "```c++" in response:
            code_pattern = r'```c\+\+((.|\n)*?)```'
        if "```Java" in response:
            code_pattern = r'```Java((.|\n)*?)```'
        if "```java" in response:
            code_pattern = r'```java((.|\n)*?)```'
        if "```Node" in response:
            code_pattern = r'```Node((.|\n)*?)```'
        if "```node" in response:
            code_pattern = r'```node((.|\n)*?)```'
        if "```Rust" in response:
            code_pattern = r'```Rust((.|\n)*?)```'
        if "```rust" in response:
            code_pattern = r'```rust((.|\n)*?)```'
        if "```PHP" in response:
            code_pattern = r'```PHP((.|\n)*?)```'
        if "```php" in response:
            code_pattern = r'```php((.|\n)*?)```'
        if "```Go" in response:
            code_pattern = r'```Go((.|\n)*?)```'
        if "```go" in response:
            code_pattern = r'```go((.|\n)*?)```'
        if "```Ruby" in response:
            code_pattern = r'```Ruby((.|\n)*?)```'
        if "```ruby" in response:
            code_pattern = r'```ruby((.|\n)*?)```'
        if "```C#" in response:
            code_pattern = r'```C#((.|\n)*?)```'
        if "```c#" in response:
            code_pattern = r'```c#((.|\n)*?)```'
        if "```csharp" in response:
            code_pattern = r'```csharp((.|\n)*?)```'

        code_blocks = re.findall(code_pattern, response, re.DOTALL)

        if type(code_blocks[-1]) == tuple or type(code_blocks[-1]) == list:
            code_str = "\n".join(code_blocks[-1])
        elif type(code_blocks[-1]) == str:
            code_str = code_blocks[-1]
        else:
            code_str = response

        return code_str

    @staticmethod
    def trim_text(text: str, trimmed_text: str):
        return text.replace(trimmed_text, '').strip()

    @staticmethod
    def replace_tag(text: str, tag: str):
        if f'<{tag}><![CDATA[' in text and f']]></{tag}>' in text:
            return text 
        else:
            return text.replace(f'<{tag}>', f'<{tag}><![CDATA[').replace(f'</{tag}>', f']]></{tag}>').strip()

    @staticmethod
    def get_sample_io_str(sample_io: any) -> str:
        if len(sample_io) > 0:
            if type(sample_io[0]) == str:
                return "\n".join(sample_io)
            if type(sample_io[0]) == dict:
                return "\n".join([f"Input:\n{io['input']}\nExpected output:\n{io['output'][0]}" for io in sample_io])
        return sample_io
    
    # append raw response to a single log file under outputs/responses/
    def log_response(self, content: str, description: str, item: dict):
        try:
            out_dir = os.path.join(os.getcwd(), "outputs", "responses")
            os.makedirs(out_dir, exist_ok=True)
            timestamp = int(time.time() * 1000)
            curtime = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime())
            file_id = item.get(self.data.id_key, timestamp) if isinstance(item, dict) else timestamp
            log_path = os.path.join(out_dir, f"MapCoder_{self.model.__class__.__name__}_responses.log")
            with open(log_path, 'a', encoding='utf-8') as fw:
                fw.write("---\n")
                fw.write(f"# timestamp: {curtime}\n")
                fw.write(f"# dataset: {self.data.__class__.__name__}\n")
                fw.write(f"# id: {file_id}\n")
                fw.write(f"# kind: {description}\n")
                fw.write(content)
                fw.write("\n\n")
        except Exception as e:
            print(f"Failed to append final code to log file: {e}", flush=True)
    
    def retrieval(self, item: dict) -> dict:
        input_kb_exemplars = [
            {
                "role": "user",
                "content": f"""Given a problem, provide relevant problems then identify the algorithm behind it and also explain the tutorial of the algorithm.
# Problem:
{self.data.get_prompt(item)}

# Exemplars:
Recall {mapping[self.k]} relevant and distinct problems (different from problem mentioned above). For each problem,
1. describe it
2. generate {self.language} code step by step to solve that problem
3. finally generate a planning to solve that problem

# Algorithm:

----------------
Important:
Your response must follow the following xml format and you can only replace the line start with # inside the tags. Make sure all tags are closed and there is a single <root> element.

<root>
<problem>
# Recall {mapping[self.k]} relevant and distinct problems (different from problem mentioned above). Write each problem in the following format.
<description>
# Describe the problem.
</description>
<code>
# Let's think step by step to solve this problem in {self.language} programming language.
</code>
<planning>
# Planning to solve this problem.
</planning>
</problem>

<problem>
# Recall {mapping[self.k]} relevant and distinct problems (different from problem mentioned above). Write each problem in the following format.
<description>
# Describe the problem.
</description>
<code>
# Let's think step by step to solve this problem in {self.language} programming language.
</code>
<planning>
# Planning to solve this problem.
</planning>
</problem>

<problem>
# Recall {mapping[self.k]} relevant and distinct problems (different from problem mentioned above). Write each problem in the following format.
<description>
# Describe the problem.
</description>
<code>
# Let's think step by step to solve this problem in {self.language} programming language.
</code>
<planning>
# Planning to solve this problem.
</planning>
</problem>

<algorithm>
# Identify the algorithm (Brute-force, Dynamic Programming, Divide-and-conquer, Greedy, Backtracking, Recursive, Binary search, and so on) that needs to be used to solve the original problem.
# Write a useful tutorial about the above mentioned algorithms. Provide a high level generic tutorial for solving this types of problem. Do not generate code.
</algorithm>
</root>
""",
            },
        ]

        print("\n\n________________________")
        print("Input for knowledge base and exemplars: ")
        print(input_kb_exemplars[0]['content'], flush=True)

        response, pr_tok_retrieval, com_tok_retrieval = self.gpt_chat(
            processed_input=input_kb_exemplars
        )
        item['api_calls'] = item.get('api_calls', 0) + 1
        self.pr_tok += pr_tok_retrieval
        self.com_tok += com_tok_retrieval

        # Post processing
        response = self.trim_text(
            response, "# Identify the algorithm (Brute-force, Dynamic Programming, Divide-and-conquer, Greedy, Backtracking, Recursive, Binary search, and so on) that needs to be used to solve the original problem.")
        response = self.trim_text(
            response, "# Write a useful tutorial about the above mentioned algorithms. Provide a high level generic tutorial for solving this types of problem. Do not generate code.")
        response = self.trim_text(
            response, "# Planning to solve this problem:")
        response = self.trim_text(
            response, f"# Let's think step by step to solve this problem in {self.language} programming language.")
        response = self.replace_tag(response, 'algorithm')
        response = self.replace_tag(response, 'description')
        response = self.replace_tag(response, 'code')
        response = self.replace_tag(response, 'planning')

        print("\n\n________________________")
        print("Response from knowledge base and exemplars: ")
        print(response, flush=True)

        # append raw response to a single log file under outputs/responses/
        self.log_response(response, "Retrieval", item)

        # parse XML with retries: if parsing fails, ask the model to re-send a strict XML-only response
        max_parse_retries = 3
        parse_attempt = 0
        parsed = None
        last_exception = None
        while parse_attempt <= max_parse_retries:
            try:
                parsed = self.parse_xml(response)
                for example_no, example in enumerate(parsed["problem"], start=1):
                    if not isinstance(example, dict):
                        raise ValueError(f"Parsed problem example {example_no} is not a dict.")
                    if 'description' not in example or 'planning' not in example:
                        raise ValueError(f"Parsed problem example {example_no} missing 'description' or 'planning' fields.")
                break
            except Exception as e:
                last_exception = e
                parse_attempt += 1
                print(f"XML parse failed on attempt {parse_attempt}: {e}", flush=True)
                if parse_attempt > max_parse_retries:
                    print("Exceeded XML parse retries. Using default parsed value and continuing.", flush=True)
                    # set a safe default parsed structure and break out to continue the workflow
                    parsed = {'problem': [{'description': '', 'planning': ''}], 'algorithm': ''}
                    break

                response_retry, pr_tok_r, com_tok_r = self.gpt_chat(
                    processed_input=input_kb_exemplars
                )
                item['api_calls'] = item.get('api_calls', 0) + 1
                self.pr_tok += pr_tok_r
                self.com_tok += com_tok_r

                # apply the same post-processing we did earlier to the new response
                response = self.trim_text(
                    response_retry, "# Identify the algorithm (Brute-force, Dynamic Programming, Divide-and-conquer, Greedy, Backtracking, Recursive, Binary search, and so on) that needs to be used to solve the original problem.")
                response = self.trim_text(
                    response, "# Write a useful tutorial about the above mentioned algorithms. Provide a high level generic tutorial for solving this types of problem. Do not generate code.")
                response = self.trim_text(
                    response, "# Planning to solve this problem:")
                response = self.trim_text(
                    response, f"# Let's think step by step to solve this problem in {self.language} programming language.")
                response = self.replace_tag(response, 'algorithm')
                response = self.replace_tag(response, 'description')
                response = self.replace_tag(response, 'code')
                response = self.replace_tag(response, 'planning')

                # log the retry response
                self.log_response(response, f"Retrieval-Retry-{parse_attempt}", item)
        
        if parse_attempt > max_parse_retries:
            parsed = {'problem': [{'description': '', 'planning': ''}], 'algorithm': ''}
        
        return parsed
    
    def planning(self, retrieval_output: dict, item: dict, algorithm_prompt: str, sample_io_prompt: str) -> list[list]:
        plannings = []

        for example_no, example in enumerate(retrieval_output["problem"], start=1):
            example_problem = example["description"]
            example_planning = example["planning"]

            input_for_problem_planning = [
                {
                    "role": "user",
                    "content": f"Given a competitive programming problem generate a concrete planning to solve the problem.\n# Problem:\n{example_problem}\n# Planning:\n{example_planning}\n{algorithm_prompt}\n## Problem to be solved:\n{self.data.get_prompt(item)}\n{sample_io_prompt}\n## Planning:\n\n----------------\nImportant: You should give only the planning to solve the problem. Do not add extra explanation or words."
                }
            ]

            print("\n\n________________________")
            print(
                f"Input for our problem planning using example: {example_no}: ")
            print(input_for_problem_planning[0]['content'], flush=True)

            planning, pr_tok_1, com_tok_1 = self.gpt_chat(
                input_for_problem_planning
            )
            item['api_calls'] += 1
            # time.sleep(1)
            self.pr_tok += pr_tok_1
            self.com_tok += com_tok_1

            # planning = self.parse_xml(planning)
            # planning['confidence'] = int(str(planning['confidence']).strip())

            print("\n\n________________________")
            print("Response from our problem planning: ")
            print(planning, flush=True)

            self.log_response(planning, f"Planning-Example-{example_no}", item)

            # input_for_planning_verification = [
            #     {
            #         "role": "user",
            #         "content": f"Given a competitive programming problem and a plan to solve the problem in {self.language}, tell whether the plan is correct to solve this problem.# Problem:\n{self.data.get_prompt(item)}\n# Planning:\n{planning}\n\n----------------\nImportant: Your response must follow the following xml format-```\n<root>\n<explanation> Discuss whether the given competitive programming problem is solvable by using the above mentioned planning.</explanation>\n<confidence> Confidence score regarding the solvability of the problem. Must be an integer between 0 and 100. </confidence>\n</root>\n```"
            #     }
            # ]

            # 只给出confidence score数字
            input_for_planning_verification = [
                {
                    "role": "user",
                    "content": f"Given a competitive programming problem and a plan to solve the problem in {self.language}, tell whether the plan is correct to solve this problem. # Problem:\n{self.data.get_prompt(item)}\n# Planning:\n{planning}\n Output: confidence score regarding the solvability of the problem\n Output Type: integer\n Output Range: 0-100\n Important: Your response must only contain the confidence score number, should not include any other explanations or words."
                }
            ]

            # Call model to get a confidence score (0-100). If the response format is invalid,
            # retry up to `max_confidence_retries` times asking the model to return strictly
            # a single integer between 0 and 100 with no extra text.
            print("Input for planning verification: ")
            print(input_for_planning_verification[0]['content'], flush=True)

            max_confidence_retries = 3
            conf_attempt = 0
            verification_score = None
            # base prompt content (we'll append stricter instruction on retries)
            verification_base = input_for_planning_verification[0]['content']

            while conf_attempt <= max_confidence_retries:
                conf_attempt += 1
                prompt_content = verification_base
                if conf_attempt > 1:
                    prompt_content += (
                        "\n\nIMPORTANT: Reply with exactly one integer between 0 and 100. "
                        "Do not include any other words, punctuation, or formatting."
                    )

                verification_input = [{"role": "user", "content": prompt_content}]

                verification_res_raw, pr_tok_1, com_tok_1 = self.gpt_chat(
                    verification_input
                )
                item['api_calls'] = item.get('api_calls', 0) + 1
                self.pr_tok += pr_tok_1
                self.com_tok += com_tok_1

                print("Response from planning verification before parsing: ")
                print(verification_res_raw, flush=True)

                # try to extract first integer from response
                try:
                    s = str(verification_res_raw).strip()
                    m = re.search(r"(-?\d+)", s)
                    if m:
                        val = int(m.group(1))
                        # clamp to 0-100
                        if val < 0:
                            val = 0
                        if val > 100:
                            val = 100
                        verification_score = val
                        print("Response from planning verification after parsing: ")
                        print(verification_score, flush=True)
                        break
                    else:
                        raise ValueError(f"No integer found in model response: {s}")
                except Exception as e:
                    print(f"Verification parse failed on attempt {conf_attempt}: {e}", flush=True)
                    # log the bad response
                    self.log_response(str(verification_res_raw), f"Verification-Retry-{conf_attempt}", item)
                    # try:
                    #     out_dir = os.path.join(os.getcwd(), "outputs", "responses")
                    #     os.makedirs(out_dir, exist_ok=True)
                    #     timestamp = int(time.time() * 1000)
                    #     file_id = item.get(self.data.id_key, timestamp) if isinstance(item, dict) else timestamp
                    #     log_path = os.path.join(out_dir, "MapCoder_responses.log")
                    #     with open(log_path, 'a', encoding='utf-8') as fw:
                    #         fw.write("---\n")
                    #         fw.write(f"# timestamp: {timestamp}\n")
                    #         fw.write(f"# dataset: {self.data.__class__.__name__}\n")
                    #         fw.write(f"# id: {file_id}\n")
                    #         fw.write(f"# kind: Verification-Retry-{conf_attempt}\n")
                    #         try:
                    #             fw.write(str(verification_res_raw))
                    #         except Exception:
                    #             fw.write(repr(verification_res_raw))
                    #         fw.write("\n\n")
                    # except Exception as e2:
                    #     print(f"Failed to append verification retry to log file: {e2}", flush=True)

                    if conf_attempt > max_confidence_retries:
                        verification_score = 100  # default to max confidence after retries

            verification_res = verification_score
            self.log_response(str(verification_res), "Verification", item)
            # try:
            #     out_dir = os.path.join(os.getcwd(), "outputs", "responses")
            #     os.makedirs(out_dir, exist_ok=True)
            #     timestamp = int(time.time() * 1000)
            #     file_id = item.get(self.data.id_key, timestamp) if isinstance(item, dict) else timestamp
            #     log_path = os.path.join(out_dir, "MapCoder_responses.log")
            #     with open(log_path, 'a', encoding='utf-8') as fw:
            #         fw.write("---\n")
            #         fw.write(f"# timestamp: {timestamp}\n")
            #         fw.write(f"# dataset: {self.data.__class__.__name__}\n")
            #         fw.write(f"# id: {file_id}\n")
            #         fw.write(f"# kind: Verification\n")
            #         try:
            #             fw.write(json.dumps(verification_res, ensure_ascii=False))
            #         except Exception:
            #             fw.write(str(verification_res))
            #         fw.write("\n\n")
            # except Exception as e:
            #     print(f"Failed to append verification_res to log file: {e}", flush=True)

            plannings.append((
                planning,
                verification_res,
                example
            ))

            # if type(self.data) == MBPPDataset and verification_res['confidence'] == 100:
            #     break

        return plannings

    def code_generation(self, plan: list, item: dict, algorithm_prompt: str, sample_io_prompt: str) -> str:
        planning, confidence, example = plan

        if type(self.data) == APPSDataset or type(self.data) == CodeContestDataset or type(self.data) == XCodeDataset:
            std_input_prompt = "## Note: Strictly follow the input and output format. The input should be taken from Standard input and output should be given to standard output. If you are writing a function then after the function definition take input using `input()` function then call the function with specified parameters and finally print the output of the function. Do not add extra print statement otherwise it will failed the test cases."
        else:
            std_input_prompt = ""

        input_for_final_code_generation = [
            {
                "role": "user",
                "content": f"Given a competitive programming problem generate {self.language} code to solve the problem.\n{algorithm_prompt}\n## Problem to be solved:\n{self.data.get_prompt(item)}\n## Planning:\n{planning}\n{sample_io_prompt}\n## Let's think step by step.\n\n----------------\nImportant:\n{std_input_prompt}\n## Your response must contain only the {self.language} code to solve this problem. Do not add extra explanation or words."
            }
        ]

        print("\n\n________________________")
        print("Input for final code generation: ")
        print(input_for_final_code_generation[0]['content'], flush=True)

        code, pr_tok_1, com_tok_1 = self.gpt_chat(
            input_for_final_code_generation
        )
        item['api_calls'] += 1
        # time.sleep(1)

        # try parsing code; if parse_code raises IndexError (empty regex matches),
        # retry calling the model up to max_code_retries times with a stricter instruction
        self.pr_tok += pr_tok_1
        self.com_tok += com_tok_1

        try:
            code = self.parse_code(code)
        except IndexError as e:
            print(f"parse_code raised IndexError: {e}. Will retry final code generation.", flush=True)
            max_code_retries = 2
            parsed_success = False
            for cretry in range(1, max_code_retries + 1):
                retry_raw, pr_tok_r, com_tok_r = self.gpt_chat(
                    input_for_final_code_generation
                )
                item['api_calls'] = item.get('api_calls', 0) + 1
                self.pr_tok += pr_tok_r
                self.com_tok += com_tok_r

                try:
                    retry_parsed = self.parse_code(retry_raw)
                    code = retry_parsed
                    parsed_success = True
                    self.log_response(retry_raw, f"final_code_retry_success-{cretry}", item)
                    break
                except Exception as e2:
                    print(f"Retry {cretry} parse_code failed: {e2}", flush=True)
                    self.log_response(retry_raw, f"final_code_retry_failed-{cretry}", item)

            if not parsed_success:
                print("Final code generation: retries exhausted, using default fallback code.", flush=True)
                lang = (self.language or "").lower()
                if 'python' in lang:
                    code = 'print("")'
                elif 'java' in lang:
                    code = 'public class Main { public static void main(String[] args) { } }'
                elif 'c++' in lang or 'cpp' in lang:
                    code = 'int main() { return 0; }'
                elif re.search(r"\bc\b", lang):
                    code = 'int main() { return 0; }'
                elif 'js' in lang or 'node' in lang or 'javascript' in lang:
                    code = 'console.log("")'
                else:
                    code = ''
                # log that we used default code
                self.log_response(code, "final_code_fallback", item)

        print("\n\n________________________")
        print("Response from final code generation: ")
        print(code, flush=True)
        self.log_response(code, "final_code", item)

        return code

    def debugging(self, plan: list, code: str, item: dict, algorithm_prompt: str) -> str:
        passed = False
        planning, _, _ = plan
        
        plan_code_response = f"## Planning: {planning}\n## Code:\n```\n{code}\n```"

        if type(self.data) == APPSDataset or type(self.data) == CodeContestDataset or type(self.data) == XCodeDataset:
            std_input_prompt = "## Note: Strictly follow the input and output format. The input should be taken from Standard input and output should be given to standard output. If you are writing a function then after the function definition take input using `input()` function then call the function with specified parameters and finally print the output of the function. Do not add extra print statement otherwise it will failed the test cases."
        else:
            std_input_prompt = ""
        
        for i in range(1, self.t + 1):
            passed, test_log = self.data.evaluate_sample_io(
                item,
                code,
                self.language
            )

            if passed:
                break

            # Use Pangu1B to analyze the failure
            pangu_input = [
                {
                    "role": "user",
                    "content": f"You are an expert programmer. The following code was generated to solve a problem but failed sample test cases.\n\n## Problem:\n{self.data.get_prompt(item)}\n\n## Generated Code:\n```\n{code}\n```\n\n## Test Report:\n{test_log}\n\nPlease analyze why the code failed and provide a specific plan to fix it. Do not generate the full code, just the analysis and fix plan."
                }
            ]
            
            print(f"Input for Pangu analysis: {i}")
            # print(qwen_input[0]['content'], flush=True)

            pangu_model = Pangu()
            analysis, q_pr_tok, q_com_tok = pangu_model.prompt(pangu_input)
            self.pr_tok += q_pr_tok
            self.com_tok += q_com_tok
            
            print(f"Pangu Analysis: {analysis}", flush=True)

            print(f"Input for improving code generation: {i}")
            input_for_improving_code = [
                {
                    "role": "user",
                    "content": f"Given a competitive programming problem you have generated {self.language} code to solve the problem. But the generated code can not pass sample test cases.\n\nHere is an analysis of the failure and a fix plan provided by an expert:\n{analysis}\n\nImprove your code to solve the problem correctly based on this analysis.\n{algorithm_prompt}\n## Problem to be solved:\n{self.data.get_prompt(item)}\n{plan_code_response}\n## Test Report:\n{test_log}\n## Modified Planning:\n## Let's think step by step to modify {self.language} Code for solving this problem.\n\n----------------\nImportant:\n{std_input_prompt}\n## Your response must contain the modified planning and then the {self.language} code inside ``` block to solve this problem."
                }
            ]

            print("\n\n________________________")
            print("Input for improving code generation: ")
            print(input_for_improving_code[0]['content'], flush=True)

            response, pr_tok_1, com_tok_1 = self.gpt_chat(
                input_for_improving_code
            )
            item['api_calls'] += 1
            # time.sleep(1)

            # try parsing code; if parse_code raises IndexError (empty regex matches),
            # retry calling the model up to max_code_retries times with a stricter instruction
            self.pr_tok += pr_tok_1
            self.com_tok += com_tok_1

            raw_code = deepcopy(code)
            try:
                code = self.parse_code(code)
            except IndexError as e:
                print(f"parse_code raised IndexError: {e}. Will retry final code generation.", flush=True)
                max_code_retries = 2
                parsed_success = False
                for cretry in range(1, max_code_retries + 1):
                    retry_raw, pr_tok_r, com_tok_r = self.gpt_chat(
                        input_for_improving_code
                    )
                    item['api_calls'] = item.get('api_calls', 0) + 1
                    self.pr_tok += pr_tok_r
                    self.com_tok += com_tok_r

                    try:
                        retry_parsed = self.parse_code(retry_raw)
                        code = retry_parsed
                        parsed_success = True
                        self.log_response(retry_raw, f"final_code_retry_success-{cretry}", item)
                        break
                    except Exception as e2:
                        print(f"Retry {cretry} parse_code failed: {e2}", flush=True)
                        self.log_response(retry_raw, f"final_code_retry_failed-{cretry}", item)

                if not parsed_success:
                    print("Final code generation: retries exhausted, using raw code.", flush=True)
                    code = raw_code
                    # log that we used raw code
                    self.log_response(code, "final_code_fallback", item)


            print("\n\n________________________")
            print("Response from improving code generation: ")
            print(response, flush=True)
            
            self.log_response(response, f"improving_code_attempt_{i}", item)

        return passed

    def run_single_pass(self, item: dict):
        print("", flush=True)

        retrieval_output = self.retrieval(item)

        algorithm_prompt = f"## Relevant Algorithm to solve the next problem:\n{ retrieval_output['algorithm']}"
        sample_io_prompt = f"## Sample Test cases: \n{self.get_sample_io_str(item['sample_io'])}\n"
        # if type(self.data) != MBPPDataset and type(self.data) != XCodeDataset else ""

        plannings = self.planning(retrieval_output, item, algorithm_prompt, sample_io_prompt)
        plannings.sort(key=lambda x: x[1], reverse=True)

        for plan in plannings:
            code = self.code_generation(plan, item, algorithm_prompt, sample_io_prompt)

            passed = self.debugging(plan, code, item, algorithm_prompt)

            if passed:
                break
        
        print("________________________\n\n", flush=True)
        return code, self.pr_tok, self.com_tok