Spaces:
Sleeping
Sleeping
File size: 13,880 Bytes
1ea26af |
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 |
#
import os
import time
import random
import re
import sys
import json
import types
import contextlib
from typing import Union, Callable
from functools import partial
import signal
import threading
import numpy as np
# rprint - simplified without colors
def rprint(inputs, style=None, timed=False):
if isinstance(inputs, str):
inputs = [inputs] # with style as the default
all_ss = []
for one_item in inputs:
if isinstance(one_item, str):
one_item = (one_item, None)
one_str, one_style = one_item # pairs
# Remove color styling - just use the string as-is
all_ss.append(one_str)
_to_print = "".join(all_ss)
if timed:
_to_print = f"[{time.ctime()}] {_to_print}"
print(_to_print)
# --
# simple adpators
zlog = rprint
zwarn = lambda x: rprint(x, style="white on red")
# --
def tuple_keys_to_str(d):
if isinstance(d, dict):
return {str(k): tuple_keys_to_str(v) for k, v in d.items()}
elif isinstance(d, list):
return [tuple_keys_to_str(i) for i in d]
else:
return d
# wrapping a function and try it multiple times
def wrapped_trying(func, default_return=None, max_times=10, wait_error_names=(), reraise=False):
# --
if max_times < 0:
return func() # directly no wrap (useful for debugging)!
# --
remaining_tryings = max_times
ret = default_return
while True:
try:
ret = func()
break # remember to jump out!!!
except Exception as e:
rprint(f"Retry with Error: {e}", style="white on red")
# Special handling for rate limit errors (429)
if type(e).__name__ == 'RateLimitError':
wait_time = 30 # Wait 30 seconds for rate limit
rprint(f"Rate limit detected, waiting {wait_time} seconds...", style="yellow")
time.sleep(wait_time)
else:
rand = random.randint(1, 5)
time.sleep(rand)
if type(e).__name__ in wait_error_names:
continue # simply wait it
else:
remaining_tryings -= 1
if remaining_tryings <= 0:
if reraise:
raise e
else:
break
return ret
# Note: GET_ENV_VAR function removed - all configuration now uses TOML-based Settings
# get until hit
def get_until_hit(d, keys, df=None):
for k in keys:
if k in d:
return d[k]
return df
# easier init with kwargs
class KwargsInitializable:
def __init__(self, _assert_existing=True, _default_init=False, **kwargs):
updates = {}
new_updates = {}
for k, v in kwargs.items():
if _assert_existing:
assert hasattr(self, k), f"Attr {k} not existing!"
v0 = getattr(self, k, None)
if v0 is not None and isinstance(v0, KwargsInitializable):
new_val = type(v0)(**v) # further make a new one!
updates[k] = f"__new__ {type(new_val)}"
elif v0 is None: # simply directly update
new_val = v
new_updates[k] = new_val
else:
new_val = type(v0)(v) # conversion
updates[k] = new_val
setattr(self, k, new_val)
# Debug output removed for clean operation
# --
# templated string (also allowing conditional prompts)
class TemplatedString:
def __init__(self, s: Union[str, Callable]):
self.str = s
def format(self, **kwargs):
if isinstance(self.str, str):
return TemplatedString.eval_fstring(self.str, **kwargs)
else: # direct call it!
return self.str(**kwargs)
@staticmethod
def eval_fstring(s: str, _globals=None, _locals=None, **kwargs):
if _locals is None:
_inner_locals = {}
else:
_inner_locals = _locals.copy()
_inner_locals.update(kwargs)
assert '"""' not in s, "Special seq not allowed!"
ret = eval('f"""'+s+'"""', _globals, _inner_locals)
return ret
# a simple wrapper class for with expression
class WithWrapper:
def __init__(self, f_start: Callable = None, f_end: Callable = None, item=None):
self.f_start = f_start
self.f_end = f_end
self.item: object = item
def __enter__(self):
if self.f_start is not None:
self.f_start()
if self.item is not None and hasattr(self.item, "__enter__"):
self.item.__enter__()
# return self if self.item is None else self.item
return self.item
def __exit__(self, exc_type, exc_val, exc_tb):
if self.item is not None and hasattr(self.item, "__exit__"):
self.item.__exit__()
if self.f_end is not None:
self.f_end()
def my_open_with(fd_or_path, mode='r', empty_std=False, **kwargs):
if empty_std and fd_or_path == '':
fd_or_path = sys.stdout if ('w' in mode) else sys.stdin
if isinstance(fd_or_path, str) and fd_or_path:
return open(fd_or_path, mode=mode, **kwargs)
else:
# assert isinstance(fd_or_path, IO)
return WithWrapper(None, None, fd_or_path)
# get unique ID
def get_unique_id(prefix=""):
import datetime
import threading
dt = datetime.datetime.now().isoformat()
ret = f"{prefix}{dt}_P{os.getpid()}_T{threading.get_native_id()}" # PID+TID
return ret
# update dict (in an incremental way)
def incr_update_dict(trg, src_dict):
for name, value in src_dict.items():
path = name.split(".")
curr = trg
for _piece in path[:-1]:
if _piece not in curr: # create one if not existing
curr[_piece] = {}
curr = curr[_piece]
_piece = path[-1]
if _piece in curr and curr[_piece] is not None:
assigning_value = type(curr[_piece])(value) # value to assign
if isinstance(assigning_value, dict) and isinstance(curr[_piece], dict):
incr_update_dict(curr[_piece], assigning_value) # further do incr
else:
curr[_piece] = assigning_value # with type conversion
else:
curr[_piece] = value # directly assign!
# --
# common response format; note: let each agent specify their own ...
# RESPONSE_FORMAT_REQUIREMENT = """## Output
# Please generate your response, your reply should strictly follow the format:
# Thought: {First, explain your reasoning for your outputs in one line.}
# Code: {Then, output your python code blob.}
# """
# parse specific formats
def parse_response(s: str, seps: list, strip=True, return_dict=False):
assert len(seps) == len(set(seps)), f"Repeated items in seps: {seps}"
ret = []
remaining_s = s
# parse them one by one
for one_sep_idx, one_sep in enumerate(seps):
try:
p1, p2 = remaining_s.split(one_sep, 1)
if p1.strip():
rprint(f"Get an unexpected piece: {p1}")
sep_val = p2
for one_sep2 in seps[one_sep_idx+1:]:
if one_sep2 in p2:
sep_val = p2.split(one_sep2, 1)[0]
break # finding one is enough!
assert p2.startswith(sep_val), "Internal error for unmatched prefix??"
remaining_s = p2[len(sep_val):]
one_val = sep_val
except: # by default None
one_val = None
ret.append(one_val)
# --
if strip:
if isinstance(strip, str):
ret = [(z.strip(strip) if isinstance(z, str) else z) for z in ret]
else:
ret = [(z.strip() if isinstance(z, str) else z) for z in ret]
if return_dict:
ret = {k: v for k, v in zip(seps, ret)}
return ret
class CodeExecutor:
def __init__(self, global_dict=None):
# self.code = code
self.results = []
self.globals = global_dict if global_dict else {}
# self.additional_imports = None
self.internal_functions = {"print": self.custom_print, "input": CodeExecutor.custom_input, "exit": CodeExecutor.custom_exit} # customized ones
self.null_stdin = False # Default to false, can be configured via settings if needed
def add_global_vars(self, **kwargs):
self.globals.update(kwargs)
@staticmethod
def extract_code(s: str):
# CODE_PATTERN = r"```(?:py[^t]|python)(.*?)```"
CODE_PATTERN = r"```(?:py[^t]|python)(.*)```" # get more codes
orig_s, hit_code = s, False
# strip _CODE_PREFIX
_CODE_PREFIX = "<|python_tag|>"
if _CODE_PREFIX in s: # strip _CODE_PREFIX
hit_code = True
_idx = s.index(_CODE_PREFIX)
s = s[_idx+len(_CODE_PREFIX):].lstrip() # strip tag
# strip all ```python ... ``` pieces
# m = re.search(r"```python(.*)```", s, flags=re.DOTALL)
if "```" in s:
hit_code = True
all_pieces = []
for piece in re.findall(CODE_PATTERN, s, flags=re.DOTALL):
all_pieces.append(piece.strip())
s = "\n".join(all_pieces)
# --
# cleaning
while s.endswith("```"): # a simple fix
s = s[:-3].strip()
ret = (s if hit_code else "")
return ret
def custom_print(self, *args):
# output = " ".join(str(arg) for arg in args)
# results.append(output)
self.results.extend(args) # note: simply adding!
@staticmethod
def custom_input(*args):
return "No input available."
@staticmethod
def custom_exit(*args):
return "Cannot exit."
def get_print_results(self, return_str=False, clear=True):
ret = self.results.copy() # a list of results
if clear:
self.results.clear()
if len(ret) == 1:
ret = ret[0] # if there is only one output
if return_str:
ret = "\n".join(ret)
return ret
def _exec(self, code, null_stdin, timeout):
original_stdin = sys.stdin # original stdin
self._timeout_flag = False
timer = None
if timeout > 0:
timer = threading.Timer(timeout, self._set_timeout_flag)
timer.start()
try:
with open(os.devnull, 'r') as fd:
if null_stdin: # change stdin
sys.stdin = fd
exec(code, self.globals) # note: no locals since things can be strange!
if self._timeout_flag:
raise TimeoutError("Code execution exceeded timeout")
finally:
if null_stdin: # change stdin
sys.stdin = original_stdin
if timer is not None:
timer.cancel() # Cancel the timer if still running
# simply remove global vars to avoid pickle errors for multiprocessing running!
# self.globals.clear() # note: simply create a new executor for each run!
def run(self, code, catch_exception=True, null_stdin=None, timeout=0):
if null_stdin is None:
null_stdin = self.null_stdin # use the default one
# --
if code: # some simple modifications
code_nopes = []
code_lines = [f"import {lib}\n" for lib in ["os", "sys"]] + ["", ""]
for one_line in code.split("\n"):
if any(re.match(r"from\s*.*\s*import\s*"+function_name, one_line.strip()) for function_name in self.globals.keys()): # no need of such imports
code_nopes.append(one_line)
else:
code_lines.append(one_line)
code = "\n".join(code_lines)
if code_nopes:
zwarn(f"Remove unneeded lines of {code_nopes}")
self.globals.update(self.internal_functions) # add internal functions
# --
if catch_exception:
try:
self._exec(code, null_stdin, timeout)
except Exception as e:
err = self.format_error(code)
# self.results.append(err)
if self.results:
err = f"{err.strip()}\n(* Partial Results={self.get_print_results()})"
if isinstance(e, TimeoutError):
err = f"{err}\n-> Please revise your code and simplify the next step to control the runtime."
self.custom_print(err) # put err
zwarn(f"Error executing code: {e}")
else:
self._exec(code, null_stdin, timeout)
# --
@staticmethod
def format_error(code: str):
import traceback
err = traceback.format_exc()
_err_line = None
_line_num = None
for _line in reversed(err.split("\n")):
ps = re.findall(r"line (\d+),", _line)
if ps:
_err_line, _line_num = _line, ps[0]
break
# print(_line_num, code.split('\n'))
try:
_line_str = code.split('\n')[int(_line_num)-1]
err = err.replace(_err_line, f"{_err_line}\n {_line_str.strip()}")
except: # if we cannot get the line
pass
return f"Code Execution Error:\n{err}"
def _set_timeout_flag(self):
self._timeout_flag = True
def get_np_generator(seed):
# Use numpy 2.0+ compatible random generator
return np.random.default_rng(seed)
# there are images in the messages
def have_images_in_messages(messages):
for message in messages:
contents = message.get("content", "")
if not isinstance(contents, list):
contents = [contents]
for one_content in contents:
if isinstance(one_content, dict):
if one_content.get("type") == "image_url":
return True
return False
|