Spaces:
Running
Running
File size: 8,692 Bytes
f398d9d |
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 |
import os
import subprocess
import openai
from agent.prompts import (
ACTION_PROMPT,
ADD_PROMPT,
COMPRESS_HISTORY_PROMPT,
LOG_PROMPT,
LOG_RESPONSE,
MODIFY_PROMPT,
PREFIX,
READ_PROMPT,
TASK_PROMPT,
UNDERSTAND_TEST_RESULTS_PROMPT,
)
from agent.utils import parse_action, parse_file_content, read_python_module_structure
VERBOSE = False
MAX_HISTORY = 100
MODEL = "gpt-3.5-turbo" # "gpt-4"
def run_gpt(
prompt_template,
stop_tokens,
max_tokens,
module_summary,
purpose,
**prompt_kwargs,
):
content = PREFIX.format(
module_summary=module_summary,
purpose=purpose,
) + prompt_template.format(**prompt_kwargs)
if VERBOSE:
print(LOG_PROMPT.format(content))
resp = openai.ChatCompletion.create(
model=MODEL,
messages=[
{"role": "system", "content": content},
],
temperature=0.0,
max_tokens=max_tokens,
stop=stop_tokens if stop_tokens else None,
)["choices"][0]["message"]["content"]
if VERBOSE:
print(LOG_RESPONSE.format(resp))
return resp
def compress_history(purpose, task, history, directory):
module_summary, _, _ = read_python_module_structure(directory)
resp = run_gpt(
COMPRESS_HISTORY_PROMPT,
stop_tokens=["observation:", "task:", "action:", "thought:"],
max_tokens=512,
module_summary=module_summary,
purpose=purpose,
task=task,
history=history,
)
history = "observation: {}\n".format(resp)
return history
def call_main(purpose, task, history, directory, action_input):
module_summary, _, _ = read_python_module_structure(directory)
resp = run_gpt(
ACTION_PROMPT,
stop_tokens=["observation:", "task:"],
max_tokens=256,
module_summary=module_summary,
purpose=purpose,
task=task,
history=history,
)
lines = resp.strip().strip("\n").split("\n")
for line in lines:
if line == "":
continue
if line.startswith("thought: "):
history += "{}\n".format(line)
elif line.startswith("action: "):
action_name, action_input = parse_action(line)
history += "{}\n".format(line)
return action_name, action_input, history, task
else:
assert False, "unknown action: {}".format(line)
return "MAIN", None, history, task
def call_test(purpose, task, history, directory, action_input):
result = subprocess.run(
["python", "-m", "pytest", "--collect-only", directory],
capture_output=True,
text=True,
)
if result.returncode != 0:
history += "observation: there are no tests! Test should be written in a test folder under {}\n".format(
directory
)
return "MAIN", None, history, task
result = subprocess.run(
["python", "-m", "pytest", directory], capture_output=True, text=True
)
if result.returncode == 0:
history += "observation: tests pass\n"
return "MAIN", None, history, task
module_summary, content, _ = read_python_module_structure(directory)
resp = run_gpt(
UNDERSTAND_TEST_RESULTS_PROMPT,
stop_tokens=[],
max_tokens=256,
module_summary=module_summary,
purpose=purpose,
task=task,
history=history,
stdout=result.stdout[:5000], # limit amount of text
stderr=result.stderr[:5000], # limit amount of text
)
history += "observation: tests failed: {}\n".format(resp)
return "MAIN", None, history, task
def call_set_task(purpose, task, history, directory, action_input):
module_summary, content, _ = read_python_module_structure(directory)
task = run_gpt(
TASK_PROMPT,
stop_tokens=[],
max_tokens=64,
module_summary=module_summary,
purpose=purpose,
task=task,
history=history,
).strip("\n")
history += "observation: task has been updated to: {}\n".format(task)
return "MAIN", None, history, task
def call_read(purpose, task, history, directory, action_input):
if not os.path.exists(action_input):
history += "observation: file does not exist\n"
return "MAIN", None, history, task
module_summary, content, _ = read_python_module_structure(directory)
f_content = (
content[action_input] if content[action_input] else "< document is empty >"
)
resp = run_gpt(
READ_PROMPT,
stop_tokens=[],
max_tokens=256,
module_summary=module_summary,
purpose=purpose,
task=task,
history=history,
file_path=action_input,
file_contents=f_content,
).strip("\n")
history += "observation: {}\n".format(resp)
return "MAIN", None, history, task
def call_modify(purpose, task, history, directory, action_input):
if not os.path.exists(action_input):
history += "observation: file does not exist\n"
return "MAIN", None, history, task
(
module_summary,
content,
_,
) = read_python_module_structure(directory)
f_content = (
content[action_input] if content[action_input] else "< document is empty >"
)
resp = run_gpt(
MODIFY_PROMPT,
stop_tokens=["action:", "thought:", "observation:"],
max_tokens=2048,
module_summary=module_summary,
purpose=purpose,
task=task,
history=history,
file_path=action_input,
file_contents=f_content,
)
new_contents, description = parse_file_content(resp)
if new_contents is None:
history += "observation: failed to modify file\n"
return "MAIN", None, history, task
with open(action_input, "w") as f:
f.write(new_contents)
history += "observation: file successfully modified\n"
history += "observation: {}\n".format(description)
return "MAIN", None, history, task
def call_add(purpose, task, history, directory, action_input):
d = os.path.dirname(action_input)
if not d.startswith(directory):
history += "observation: files must be under directory {}\n".format(directory)
elif not action_input.endswith(".py"):
history += "observation: can only write .py files\n"
else:
if d and not os.path.exists(d):
os.makedirs(d)
if not os.path.exists(action_input):
module_summary, _, _ = read_python_module_structure(directory)
resp = run_gpt(
ADD_PROMPT,
stop_tokens=["action:", "thought:", "observation:"],
max_tokens=2048,
module_summary=module_summary,
purpose=purpose,
task=task,
history=history,
file_path=action_input,
)
new_contents, description = parse_file_content(resp)
if new_contents is None:
history += "observation: failed to write file\n"
return "MAIN", None, history, task
with open(action_input, "w") as f:
f.write(new_contents)
history += "observation: file successfully written\n"
history += "obsertation: {}\n".format(description)
else:
history += "observation: file already exists\n"
return "MAIN", None, history, task
NAME_TO_FUNC = {
"MAIN": call_main,
"UPDATE-TASK": call_set_task,
"MODIFY-FILE": call_modify,
"READ-FILE": call_read,
"ADD-FILE": call_add,
"TEST": call_test,
}
def run_action(purpose, task, history, directory, action_name, action_input):
if action_name == "COMPLETE":
exit(0)
# compress the history when it is long
if len(history.split("\n")) > MAX_HISTORY:
if VERBOSE:
print("COMPRESSING HISTORY")
history = compress_history(purpose, task, history, directory)
assert action_name in NAME_TO_FUNC
print("RUN: ", action_name, action_input)
return NAME_TO_FUNC[action_name](purpose, task, history, directory, action_input)
def run(purpose, directory, task=None):
history = ""
action_name = "UPDATE-TASK" if task is None else "MAIN"
action_input = None
while True:
print("")
print("")
print("---")
print("purpose:", purpose)
print("task:", task)
print("---")
print(history)
print("---")
action_name, action_input, history, task = run_action(
purpose,
task,
history,
directory,
action_name,
action_input,
) |