File size: 19,606 Bytes
a80200a |
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 |
import os
def apply_RL_prompt(chunk, args, budget):
if args.prompt_type == "deepseek3" and os.environ['tip'] == "Ahead":
#### Ahead方法为:在问题的末尾加上remaining。
return RL_deepseek3_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "prompt":
return prompt_format(chunk, budget)
elif args.prompt_type == "qwen25-math-cot" and os.environ['tip'] == "prompt":
return prompt_format_qw(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "prompt-based":
return deepseek3_prompt_based(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "prompt-based1":
return deepseek3_prompt_based1(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "prompt-based2":
return deepseek3_prompt_based2(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "prompt-based3":
return deepseek3_prompt_based3(chunk, budget)
elif os.environ['tip'] == "default":
return chunk
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "ATD_A" or os.environ['tip'] == "ATD_R":
return ATD_A_deepseek3_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "TCM":
return TCM_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "TCMv2":
return TCMv2_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "withoutremaining":
return withoutremaining_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "8ratio":
return _8ratio_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and "prompt_v1" in os.environ['tip']:
return prompt_v1_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and "prompt_v2" in os.environ['tip']:
return prompt_v2_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and "prompt_cod" in os.environ['tip']:
return prompt_cod_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and "prompt_c2f" in os.environ['tip']:
return prompt_c2f_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "Laser":
return Laser_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "TCM+":
return TCMv2_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "SST":
return SST_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "Tokenskip":
return Tokenskip_prompt(chunk, budget)
elif args.prompt_type == "deepseek3" and os.environ['tip'] == "thinkprune":
return thinkprune_prompt(chunk, budget)
else:
return chunk
def Tokenskip_prompt(chunk, budget):
find_strings = "<|Assistant|>"
alpha = os.environ['alpha']
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
add_prompt = f'<|end▁of▁sentence|>{alpha}<|end▁of▁sentence|>'
chunk[i] = head + add_prompt + find_strings + tail
return chunk
def SST_prompt(chunk, budget):
find_strings = "<|Assistant|>"
up_to_50 = budget // 50 + 1
N = up_to_50
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)\n<remaining>{budget}</remaining>\n'
add_prompt = f"\n(Complete thinking within {N} <countdown> or fewer.)"
# add_prompt = f'\n<remaining>{budget}</remaining>\n'
add_response = "\n<think>\n\n<countdown>\n"
# head += f"\n<remaining>{budget}</remaining>\n"
chunk[i] = head + add_prompt + find_strings + tail + add_response
# print(f"chunk[i] = {chunk[i]}")
return chunk
def TCMv2_prompt(chunk, budget):
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)\n<remaining>{budget}</remaining>\n'
add_prompt = f"\n(Complete thinking within \n<remaining>{budget}</remaining>\n tokens or fewer.)"
# add_prompt = f'\n<remaining>{budget}</remaining>\n'
add_response = f""
# head += f"\n<remaining>{budget}</remaining>\n"
chunk[i] = head + add_prompt + find_strings + add_response + tail
# print(f"chunk[i] = {chunk[i]}")
return chunk
def withoutremaining_prompt(chunk, budget):
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)\n<remaining>{budget}</remaining>\n'
add_prompt = f"\n(Complete thinking within {budget} tokens or fewer.)"
# add_prompt = f'\n<remaining>{budget}</remaining>\n'
add_response = f""
# head += f"\n<remaining>{budget}</remaining>\n"
chunk[i] = head + add_prompt + find_strings + add_response + tail
# print(f"chunk[i] = {chunk[i]}")
return chunk
def _8ratio_prompt(chunk, budget):
os.environ['budget'] = str(budget)
print(f"budget = {budget}")
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
add_prompt = f"\n(Complete thinking within {budget} tokens or fewer, 7 special tokens ( \n<remaining>7/8</remaining>\n , \n<remaining>6/8</remaining>\n , \n<remaining>5/8</remaining>\n , \n<remaining>4/8</remaining>\n , \n<remaining>3/8</remaining>\n , \n<remaining>2/8</remaining>\n , \n<remaining>1/8</remaining>\n ) will split the thinking process into 8 parts.)"
add_response = f""
chunk[i] = head + add_prompt + find_strings + add_response + tail
return chunk
def Laser_prompt(chunk, budget):
os.environ['budget'] = str(budget)
print(f"budget = {budget}")
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
add_prompt = "Please reason step by step, and put your final answer within \\boxed{}"
add_response = f""
chunk[i] = head + add_prompt + find_strings + add_response + tail
return chunk
def prompt_v1_prompt(chunk, budget):
os.environ['budget'] = str(budget)
print(f"budget = {budget}")
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
add_prompt = f"\n(Complete thinking within {budget} tokens or fewer, please output the remaining number of tokens every 200 tokens to facilitate control of the remaining length of the thinking process, here is a template: 'now remaining tokens: xxx', xxx is the real remaining number of tokens.)"
add_response = f""
chunk[i] = head + add_prompt + find_strings + add_response + tail
return chunk
def prompt_v2_prompt(chunk, budget):
os.environ['budget'] = str(budget)
print(f"budget = {budget}")
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
add_prompt = f"\n(Complete thinking within {budget} tokens or fewer)"
add_response = f""
chunk[i] = head + add_prompt + find_strings + add_response + tail
return chunk
def prompt_cod_prompt(chunk, budget):
os.environ['budget'] = str(budget)
print(f"budget = {budget}")
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
add_prompt = f'''
Please solve the following math problem step by step, but only record the most essential steps during the reasoning process. Each step should be concise and clear, focusing directly on the core calculation or logical transformation. No detailed explanation is needed; just write down the key derivations. After completing the reasoning, provide the final answer after.
Example:
Problem: Jason has 20 lollipops. After giving some to Denny, he now has 12 left. How many did he give to Denny?
Reasoning:
20 - x = 12
x = 20 - 12
x = 8
8
Now please solve the above problem.\n(Complete thinking within {budget} tokens or fewer)
'''
add_response = f""
chunk[i] = head + add_prompt + find_strings + add_response + tail
return chunk
def prompt_c2f_prompt(chunk, budget):
os.environ['budget'] = str(budget)
print(f"budget = {budget}")
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
add_prompt = f'''
Use the following pattern to solve the problem:
**Coarse-Grained Reasoning**
Provide a brief analysis and initial answer, focusing
on efficiency and conciseness.
**Fine-Grained Reasoning**
Provide detailed reasoning step by step and a refined
answer, focusing on correctness and rigor.
Conclude with:
Therefore, the final answer is: \\boxed{{[answer]}}.
Where [answer] is just the final number or expression
that solves the problem.
Now please solve the above problem.\n(Complete thinking within {budget} tokens or fewer)
'''
add_response = f""
chunk[i] = head + add_prompt + find_strings + add_response + tail
return chunk
def thinkprune_prompt(chunk, budget):
SYSTEM_PROMPT = (
"A conversation between User and Assistant. The user asks a question, "
"and the Assistant solves it."
" The assistant first thinks about the reasoning process in the mind and then provides the user"
" with the answer. The reasoning process and answer are enclosed within <think></think> and"
" <answer></answer> tags, respectively, i.e., <think> reasoning process here</think>"
"<answer> answer here</answer>."
)
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)\n<remaining>{budget}</remaining>\n'
add_prompt = SYSTEM_PROMPT + f"The output of the assistant should be within {budget} tokens"
# add_prompt = f'\n<remaining>{budget}</remaining>\n'
add_response = f""
# head += f"\n<remaining>{budget}</remaining>\n"
chunk[i] = head + add_prompt + find_strings + add_response + tail
# print(f"chunk[i] = {chunk[i]}")
return chunk
def TCM_prompt(chunk, budget):
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)'
add_prompt = f"\n(Complete thinking within \n<remaining>{budget}</remaining>\n tokens or fewer.)"
# add_prompt = f'\n(Complete thinking within {budget} tokens or fewer.)\n<remaining>{budget}</remaining>\n'
# add_prompt = f'\n<remaining>{budget}</remaining>\n'
add_response = f""
# head += f"\n<remaining>{budget}</remaining>\n"
chunk[i] = head + add_prompt + find_strings + add_response + tail
# print(f"chunk[i] = {chunk[i]}")
return chunk
def prompt_format(chunk, budget):
'''
<|User|>Convert the point $(0,3)$ in rectangular coordinates to polar coordinates.
Enter your answer in the form $(r,\theta),$ where $r > 0$ and $0 \le \theta < 2 \pi.$
<|Assistant|>
'''
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
head += f"\n(Complete thinking within {budget} tokens or fewer.)\n"
chunk[i] = head + find_strings + tail
return chunk
def prompt_format_qw(chunk, budget):
'''
<|User|>Convert the point $(0,3)$ in rectangular coordinates to polar coordinates.
Enter your answer in the form $(r,\theta),$ where $r > 0$ and $0 \le \theta < 2 \pi.$
<|Assistant|>
'''
find_strings = "<|im_start|>assistant"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
head += f"\n(Complete thinking within {budget} tokens or fewer.)\n"
chunk[i] = head + find_strings + tail
return chunk
def ATD_A_deepseek3_prompt(chunk, budget):
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
add_prompt = f'\n(Respond in {budget} tokens or fewer. Complete the process between <think> and </think> within the token budget. Display the countdown exponentially as <remaining>xxx</remaining>, where xxx = 50 * 2^n, n >= 0. Think more concisely as countdown decreases.)\n'
add_response = f"\n(I will complete the process within {budget} tokens and show the countdown as <remaining>xxx</remaining>, following the exponential rule.I will think more concisely as countdown decreases.)\n"
# head += f"\n<remaining>{budget}</remaining>\n"
chunk[i] = head + add_prompt + find_strings + add_response + tail
return chunk
def RL_deepseek3_prompt(chunk, budget):
'''
<|User|>Convert the point $(0,3)$ in rectangular coordinates to polar coordinates.
Enter your answer in the form $(r,\theta),$ where $r > 0$ and $0 \le \theta < 2 \pi.$
<|Assistant|>
'''
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
head += f"(Please respond in {budget} tokens or fewer)\n"
# head += f"\n<remaining>{budget}</remaining>\n"
chunk[i] = head + find_strings + tail
return chunk
def deepseek3_prompt_based(chunk, budget):
'''
<|User|>Convert the point $(0,3)$ in rectangular coordinates to polar coordinates.
Enter your answer in the form $(r,\theta),$ where $r > 0$ and $0 \le \theta < 2 \pi.$
<|Assistant|>
'''
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
head += f"You should finish thinking with in {budget} tokens.\n"
chunk[i] = head + find_strings + tail
return chunk
def deepseek3_prompt_based1(chunk, budget):
'''
<|User|>Convert the point $(0,3)$ in rectangular coordinates to polar coordinates.
Enter your answer in the form $(r,\theta),$ where $r > 0$ and $0 \le \theta < 2 \pi.$
<|Assistant|>
'''
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
head += f"(Please respond in {budget} tokens or fewer)\n"
chunk[i] = head + find_strings + tail
return chunk
def deepseek3_prompt_based2(chunk, budget):
'''
<|User|>Convert the point $(0,3)$ in rectangular coordinates to polar coordinates.
Enter your answer in the form $(r,\theta),$ where $r > 0$ and $0 \le \theta < 2 \pi.$
<|Assistant|>
'''
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
head += f"(HARD STOP at {budget} tokens)\n"
chunk[i] = head + find_strings + tail
return chunk
def deepseek3_prompt_based3(chunk, budget):
'''
<|User|>Convert the point $(0,3)$ in rectangular coordinates to polar coordinates.
Enter your answer in the form $(r,\theta),$ where $r > 0$ and $0 \le \theta < 2 \pi.$
<|Assistant|>
'''
find_strings = "<|Assistant|>"
for i in range(len(chunk)):
head = chunk[i].split(find_strings)[0]
tail = chunk[i].split(find_strings)[1]
head += f"(Response length limit: Strictly {budget} tokens max. Budget cannot be exceeded)\n"
chunk[i] = head + find_strings + tail
return chunk
# def solve_final_answer(chunk):
# k = 0
# for i in range(len(chunk)):
# if "**Final Answer**\\boxed" in chunk[i][:-10] and "<|end▁of▁sentence|>" not in chunk[i]:
# chunk[i] += "<|end▁of▁sentence|>"
# k += 1
# print(f"###added {k} final answer!")
# return chunk
# import re
def is_balanced(s: str) -> bool:
"""验证大括号是否成对且正确嵌套"""
stack = 0
for char in s:
if char == "{":
stack += 1
elif char == "}":
stack -= 1
if stack < 0:
return False
return stack == 0
def solve_final_answer(chunk: list) -> list:
"""处理包含嵌套大括号的答案匹配"""
end_chunk = []
open_chunk = []
k = 0
pattern = "**Final Answer**\\boxed{" # 初级匹配
for i in range(len(chunk)):
line = chunk[i]
if not pattern in line:
open_chunk.append(chunk[i])
continue
# 深度扫描完整结构
start_idx = line.find('**Final Answer**\\boxed{')
if start_idx == -1:
open_chunk.append(chunk[i])
continue
# 手动解析嵌套结构
stack = 1
end_idx = start_idx + len('**Final Answer**\\boxed{')
while end_idx < len(line) and stack > 0:
if line[end_idx] == "{":
stack += 1
elif line[end_idx] == "}":
stack -= 1
end_idx += 1
# 验证闭合状态
if stack == 0 and is_balanced(line[start_idx:end_idx]):
chunk[i] += "<|end▁of▁sentence|>" # 保持原有操作
k += 1
end_chunk.append(chunk[i])
else:
open_chunk.append(chunk[i])
print(f"### Find {k} anwsers have final answer!")
return chunk, end_chunk, open_chunk |