File size: 15,116 Bytes
8496edd |
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 |
from .base_agent import BaseAgent
from prompt.constants import modeling_methods
from prompt.template import (TASK_ANALYSIS_PROMPT, TASK_RESULT_PROMPT, TASK_ANSWER_PROMPT,
TASK_FORMULAS_PROMPT, TASK_FORMULAS_CRITIQUE_PROMPT, TASK_FORMULAS_IMPROVEMENT_PROMPT,
TASK_MODELING_PROMPT, TASK_MODELING_CRITIQUE_PROMPT, TASK_MODELING_IMPROVEMENT_PROMPT,
TASK_CODING_PROMPT, TASK_CODING_DEBUG_PROMPT, CODE_STRUCTURE_PROMPT,
TASK_RESULT_WITH_CODE_PROMPT, COO_PROMPT, TASK_CODING_WO_COO_PROMPT)
import sys
import os
import subprocess
import selectors
import tiktoken
import json
class EnvException(Exception):
def __init__(self, message):
self.message = message
def __str__(self):
return self.message
def execute_script(script_path, work_dir):
try:
device = 0
python = "python"
cmd = f"CUDA_VISIBLE_DEVICES={device} {python} -u {script_path}"
process = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, shell=True, cwd=work_dir)
stdout_lines = []
stderr_lines = []
selector = selectors.DefaultSelector()
selector.register(process.stdout, selectors.EVENT_READ)
selector.register(process.stderr, selectors.EVENT_READ)
while process.poll() is None and selector.get_map():
events = selector.select(timeout=1)
for key, _ in events:
line = key.fileobj.readline()
if key.fileobj == process.stdout:
print("STDOUT:", line, end =" ")
stdout_lines.append(line)
else:
print("STDERR:", line, end =" ")
stderr_lines.append(line)
for line in process.stdout:
line = line
print("STDOUT:", line, end =" ")
stdout_lines.append(line)
for line in process.stderr:
line = line
print("STDERR:", line, end =" ")
stderr_lines.append(line)
return_code = process.returncode
if return_code != 0:
observation = "".join(stderr_lines)
else:
observation = "".join(stdout_lines)
if observation == "" and return_code == 0:
# printed to stderr only
observation = "".join(stderr_lines)
return "The script has been executed. Here is the output:\n" + observation
except Exception as e:
print("++++", "Wrong!")
raise EnvException(f"Something went wrong in executing {script_path}: {e}. Please check if it is ready to be executed.")
class Task(BaseAgent):
def __init__(self, llm, coo=True, rag=True):
super().__init__(llm)
self.coo = coo
self.rag = rag
if coo:
self.coo_prompt = COO_PROMPT
else:
self.coo_prompt = ""
def analysis(self, prompt: str, task_description: str, user_prompt: str = ''):
prompt = TASK_ANALYSIS_PROMPT.format(prompt=prompt, coo_prompt=self.coo_prompt, task_description=task_description, user_prompt=user_prompt).strip()
return self.llm.generate(prompt)
def formulas_actor(self, prompt: str, data_summary: str, task_description: str, task_analysis: str, modeling_methods: str, user_prompt: str = ''):
prompt = TASK_FORMULAS_PROMPT.format(prompt=prompt, coo_prompt=self.coo_prompt, data_summary=data_summary, task_description=task_description, task_analysis=task_analysis, modeling_methods=modeling_methods, user_prompt=user_prompt).strip()
return self.llm.generate(prompt)
def formulas_critic(self, data_summary: str, task_description: str, task_analysis: str, modeling_formulas: str):
prompt = TASK_FORMULAS_CRITIQUE_PROMPT.format(data_summary=data_summary, task_description=task_description, task_analysis=task_analysis, modeling_formulas=modeling_formulas).strip()
return self.llm.generate(prompt)
def formulas_improvement(self, data_summary: str, task_description: str, task_analysis: str, modeling_formulas: str, modeling_formulas_critique: str, user_prompt: str = ''):
prompt = TASK_FORMULAS_IMPROVEMENT_PROMPT.format(data_summary=data_summary, task_description=task_description, task_analysis=task_analysis, modeling_formulas=modeling_formulas, modeling_formulas_critique=modeling_formulas_critique, user_prompt=user_prompt).strip()
return self.llm.generate(prompt)
def formulas(self, prompt: str, data_summary: str, task_description: str, task_analysis: str, modeling_methods: str, round: int = 1, user_prompt: str = ''):
formulas = self.formulas_actor(prompt, data_summary, task_description, task_analysis, modeling_methods, user_prompt)
if self.rag:
for i in range(round):
print(f'FORMULAS Round {i+1}')
formulas_critique = self.formulas_critic(data_summary, task_description, task_analysis, formulas)
formulas = self.formulas_improvement(data_summary, task_description, task_analysis, formulas, formulas_critique, user_prompt)
return formulas
def modeling_actor(self, prompt: str, data_summary: str, task_description: str, task_analysis: str, formulas: str, user_prompt: str = ''):
prompt = TASK_MODELING_PROMPT.format(prompt=prompt, coo_prompt=self.coo_prompt, data_summary=data_summary, task_description=task_description, task_analysis=task_analysis, modeling_formulas=formulas, user_prompt=user_prompt).strip()
return self.llm.generate(prompt)
# def modeling_critic(self, task_description: str, task_analysis: str, data_summary: str, formulas: str, modeling_process: str):
# prompt = TASK_MODELING_CRITIQUE_PROMPT.format(task_description=task_description, task_analysis=task_analysis, data_summary=data_summary, modeling_formulas=formulas, modeling_process=modeling_process).strip()
# return self.llm.generate(prompt)
# def modeling_improvement(self, task_description: str, task_analysis: str, data_summary: str, formulas: str, modeling_process: str, modeling_process_critique: str):
# prompt = TASK_MODELING_IMPROVEMENT_PROMPT.format(task_description=task_description, task_analysis=task_analysis, data_summary=data_summary, modeling_formulas=formulas, modeling_process=modeling_process, modeling_process_critique=modeling_process_critique).strip()
# return self.llm.generate(prompt)
# def modeling(self, task_description: str, task_analysis: str, data_summary: str, formulas: str, round: int = 1):
# process = self.modeling_actor(task_description, task_analysis, data_summary, formulas)
# for i in range(round):
# print(f'MODELING Round {i+1}')
# process_critique = self.modeling_critic(task_description, task_analysis, data_summary, formulas, process)
# process = self.modeling_improvement(task_description, task_analysis, data_summary, formulas, process, process_critique)
# return process
def modeling(self, prompt: str, data_summary: str, task_description: str, task_analysis: str, formulas: str, round: int = 1, user_prompt: str = ''):
return self.modeling_actor(prompt, data_summary, task_description, task_analysis, formulas, user_prompt)
def modeling_actor(self, prompt: str, data_summary: str, task_description: str, task_analysis: str, formulas: str, modeling: str, user_prompt: str = ''):
prompt = TASK_MODELING_PROMPT.format(prompt=prompt, coo_prompt=self.coo_prompt, data_summary=data_summary, task_description=task_description, task_analysis=task_analysis, modeling_formulas=formulas, modeling_methods=modeling, user_prompt=user_prompt).strip()
return self.llm.generate(prompt)
def coding_actor(self, data_file, data_summary, variable_description, task_description: str, task_analysis: str, formulas: str, modeling: str, dependent_file_prompt: str, code_template: str, script_name: str, work_dir: str, user_prompt: str = ''):
if self.coo:
prompt = TASK_CODING_PROMPT.format(data_file=data_file, data_summary=data_summary, variable_description=variable_description, task_description=task_description, task_analysis=task_analysis, modeling_formulas=formulas, modeling_process=modeling, dependent_file_prompt=dependent_file_prompt, code_template=code_template, user_prompt=user_prompt).strip()
else:
prompt = TASK_CODING_WO_COO_PROMPT.format(data_file=data_file, data_summary=data_summary, variable_description=variable_description, task_description=task_description, task_analysis=task_analysis, modeling_formulas=formulas, modeling_process=modeling, code_template=code_template, user_prompt=user_prompt).strip()
max_retry = 0
while max_retry < 5:
max_retry += 1
try:
completion = self.llm.generate(prompt)
new_content = completion.split("```python")[1].split("```")[0].strip()
break
except Exception as e:
# Format control.
print(f"Retry! The code does not start with ```python")
continue
with open(os.path.join(work_dir, script_name), "w") as f:
f.write(new_content)
# Execute the script.
try:
observation = execute_script(script_name, work_dir)
## If observation is too long, we only keep the last ~2k tokens.
enc = tiktoken.get_encoding("cl100k_base")
tokens = len(enc.encode(observation))
if tokens >= 2000:
observation = observation[:2000]
tokens = len(enc.encode(observation))
except Exception as e:
print(e)
input("Ah oh, Got stuck! Press any key to continue.")
return new_content, observation
def coding_debugger(self, code_template: str, modeling: str, code: str, observation: str, script_name: str, work_dir: str, user_prompt: str = ''):
prompt = TASK_CODING_DEBUG_PROMPT.format(code_template=code_template, modeling_process=modeling, code=code, observation=observation, user_prompt=user_prompt).strip()
max_retry = 0
while max_retry < 5:
max_retry += 1
try:
completion = self.llm.generate(prompt)
new_content = completion.split("```python")[1].split("```")[0].strip()
break
except Exception as e:
# Format control.
print(f"Retry! The code does not start with ```python")
continue
with open(os.path.join(work_dir, script_name), "w") as f:
f.write(new_content)
# Execute the script.
try:
observation = execute_script(script_name, work_dir)
## If observation is too long, we only keep the last ~2k tokens.
enc = tiktoken.get_encoding("cl100k_base")
tokens = len(enc.encode(observation))
if tokens >= 2000:
observation = observation[:2000]
tokens = len(enc.encode(observation))
except Exception as e:
print(e)
input("Ah oh, Got stuck! Press any key to continue.")
return new_content, observation
def coding(self, data_file, data_summary, variable_description, task_description: str, task_analysis: str, formulas: str, modeling: str, dependent_file_prompt: str, code_template: str, script_name: str, work_dir: str, try_num: int = 5, round: int = 1, user_prompt: str = ''):
for i in range(try_num):
print("="*10 + f" Try: {i + 1} " + "="*10)
iteration = 0
max_iteration = 3
while iteration < max_iteration:
print("="*10 + f" Iteration: {iteration + 1} " + "="*10)
if iteration == 0:
code, observation = self.coding_actor(data_file, data_summary, variable_description, task_description, task_analysis, formulas, modeling, dependent_file_prompt, code_template, script_name, work_dir, user_prompt)
# If the script has been successfully executed: Exit.
if "Traceback (most recent call last):" not in observation and "SyntaxError: invalid syntax" not in observation and "IndentationError" not in observation:
return code, True, observation.split("The script has been executed. Here is the output:\n")[1]
else:
code, observation = self.coding_debugger(code_template, modeling, code, observation, script_name, work_dir, user_prompt)
# If the script has been successfully executed: Exit.
if "Traceback (most recent call last):" not in observation and "SyntaxError: invalid syntax" not in observation and "IndentationError" not in observation:
return code, True, observation.split("The script has been executed. Here is the output:\n")[1]
iteration += 1
return code, False, None
def result(self, task_description: str, task_analysis: str, task_formulas: str, task_modeling: str, user_prompt: str = '', execution_result: str = ''):
if execution_result == '':
prompt = TASK_RESULT_PROMPT.format(task_description=task_description, task_analysis=task_analysis, task_formulas=task_formulas, task_modeling=task_modeling, user_prompt=user_prompt).strip()
else:
prompt = TASK_RESULT_WITH_CODE_PROMPT.format(task_description=task_description, task_analysis=task_analysis, task_formulas=task_formulas, task_modeling=task_modeling, user_prompt=user_prompt, execution_result=execution_result).strip()
return self.llm.generate(prompt)
def answer(self, task_description: str, task_analysis: str, task_formulas: str, task_modeling: str, task_result: str, user_prompt: str = ''):
prompt = TASK_ANSWER_PROMPT.format(task_description=task_description, task_analysis=task_analysis, task_formulas=task_formulas, task_modeling=task_modeling, task_result=task_result, user_prompt=user_prompt).strip()
return self.llm.generate(prompt)
def extract_code_structure(self, task_id, code: str, save_path: str):
prompt = CODE_STRUCTURE_PROMPT.format(code=code, save_path=save_path)
count = 0
for i in range(5):
try:
strucutre = self.llm.generate(prompt)
structure_string = strucutre.strip('```json\n').strip('```')
structure_json = json.loads(structure_string)
for i in range(len(structure_json['file_outputs'])):
structure_json['file_outputs'][i]['file_description'] = 'This file is generated by code for Task {}. '.format(task_id) + structure_json['file_outputs'][i]['file_description']
return structure_json
except:
continue
if count == 5:
sys.exit("Fail at extract_code_structure")
|