File size: 9,148 Bytes
7919c19
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import subprocess
#import openai
from huggingface_hub import InferenceClient
import gradio as gr
import prompts
client = InferenceClient(
    "mistralai/Mixtral-8x7B-Instruct-v0.1"
)

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))
    
    
    #formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", history)
    stream = client.text_generation({"role": "system", "content": content}, **generate_kwargs, stream=True, details=True, return_full_text=False)["choices"][0]["message"]["content"]
    output = ""
    
    '''
    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,
        )