Spaces:
Running
on
A10G
Running
on
A10G
Upload octotools_engine_openai.py
Browse files
octotools/engine/octotools_engine_openai.py
ADDED
@@ -0,0 +1,269 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
try:
|
2 |
+
from openai import OpenAI
|
3 |
+
except ImportError:
|
4 |
+
raise ImportError("If you'd like to use OpenAI models, please install the openai package by running `pip install openai`, and add 'OPENAI_API_KEY' to your environment variables.")
|
5 |
+
|
6 |
+
import os
|
7 |
+
import json
|
8 |
+
import base64
|
9 |
+
import platformdirs
|
10 |
+
from tenacity import (
|
11 |
+
retry,
|
12 |
+
stop_after_attempt,
|
13 |
+
wait_random_exponential,
|
14 |
+
)
|
15 |
+
from typing import List, Union
|
16 |
+
|
17 |
+
from .base import EngineLM, CachedEngine
|
18 |
+
|
19 |
+
import openai
|
20 |
+
|
21 |
+
from dotenv import load_dotenv
|
22 |
+
load_dotenv()
|
23 |
+
|
24 |
+
# Define global constant for structured models
|
25 |
+
# https://platform.openai.com/docs/guides/structured-outputs
|
26 |
+
# https://cookbook.openai.com/examples/structured_outputs_intro
|
27 |
+
from pydantic import BaseModel
|
28 |
+
|
29 |
+
class DefaultFormat(BaseModel):
|
30 |
+
response: str
|
31 |
+
|
32 |
+
# Define global constant for structured models
|
33 |
+
OPENAI_STRUCTURED_MODELS = ['gpt-4o', 'gpt-4o-2024-08-06','gpt-4o-mini', 'gpt-4o-mini-2024-07-18','deepseek']
|
34 |
+
|
35 |
+
|
36 |
+
class ChatOpenAI(EngineLM, CachedEngine):
|
37 |
+
DEFAULT_SYSTEM_PROMPT = "You are a helpful, creative, and smart assistant."
|
38 |
+
|
39 |
+
def __init__(
|
40 |
+
self,
|
41 |
+
model_string="gpt-4o-mini-2024-07-18",
|
42 |
+
system_prompt=DEFAULT_SYSTEM_PROMPT,
|
43 |
+
is_multimodal: bool=False,
|
44 |
+
# enable_cache: bool=True,
|
45 |
+
enable_cache: bool=False, # NOTE: disable cache for now
|
46 |
+
api_key: str=None,
|
47 |
+
**kwargs):
|
48 |
+
"""
|
49 |
+
:param model_string:
|
50 |
+
:param system_prompt:
|
51 |
+
:param is_multimodal:
|
52 |
+
"""
|
53 |
+
if enable_cache:
|
54 |
+
root = platformdirs.user_cache_dir("octotools")
|
55 |
+
cache_path = os.path.join(root, f"cache_openai_{model_string}.db")
|
56 |
+
# For example, cache_path = /root/.cache/octotools/cache_openai_gpt-4o-mini.db
|
57 |
+
# print(f"Cache path: {cache_path}")
|
58 |
+
|
59 |
+
self.image_cache_dir = os.path.join(root, "image_cache")
|
60 |
+
os.makedirs(self.image_cache_dir, exist_ok=True)
|
61 |
+
|
62 |
+
super().__init__(cache_path=cache_path)
|
63 |
+
|
64 |
+
self.system_prompt = system_prompt
|
65 |
+
if api_key is None:
|
66 |
+
raise ValueError("Please set the OPENAI_API_KEY environment variable if you'd like to use OpenAI models.")
|
67 |
+
|
68 |
+
self.client = OpenAI(
|
69 |
+
api_key=api_key,
|
70 |
+
)
|
71 |
+
self.model_string = model_string
|
72 |
+
self.is_multimodal = is_multimodal
|
73 |
+
self.enable_cache = enable_cache
|
74 |
+
|
75 |
+
if enable_cache:
|
76 |
+
print(f"!! Cache enabled for model: {self.model_string}")
|
77 |
+
else:
|
78 |
+
print(f"!! Cache disabled for model: {self.model_string}")
|
79 |
+
|
80 |
+
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(5))
|
81 |
+
def generate(self, content: Union[str, List[Union[str, bytes]]], system_prompt=None, **kwargs):
|
82 |
+
try:
|
83 |
+
# Print retry attempt information
|
84 |
+
attempt_number = self.generate.retry.statistics.get('attempt_number', 0) + 1
|
85 |
+
if attempt_number > 1:
|
86 |
+
print(f"Attempt {attempt_number} of 5")
|
87 |
+
|
88 |
+
if isinstance(content, str):
|
89 |
+
return self._generate_text(content, system_prompt=system_prompt, **kwargs)
|
90 |
+
|
91 |
+
elif isinstance(content, list):
|
92 |
+
if (not self.is_multimodal):
|
93 |
+
raise NotImplementedError("Multimodal generation is only supported for GPT-4 models.")
|
94 |
+
|
95 |
+
return self._generate_multimodal(content, system_prompt=system_prompt, **kwargs)
|
96 |
+
|
97 |
+
except openai.LengthFinishReasonError as e:
|
98 |
+
print(f"Token limit exceeded: {str(e)}")
|
99 |
+
print(f"Tokens used - Completion: {e.completion.usage.completion_tokens}, Prompt: {e.completion.usage.prompt_tokens}, Total: {e.completion.usage.total_tokens}")
|
100 |
+
return {
|
101 |
+
"error": "token_limit_exceeded",
|
102 |
+
"message": str(e),
|
103 |
+
"details": {
|
104 |
+
"completion_tokens": e.completion.usage.completion_tokens,
|
105 |
+
"prompt_tokens": e.completion.usage.prompt_tokens,
|
106 |
+
"total_tokens": e.completion.usage.total_tokens
|
107 |
+
}
|
108 |
+
}
|
109 |
+
except openai.RateLimitError as e:
|
110 |
+
print(f"Rate limit error encountered: {str(e)}")
|
111 |
+
return {
|
112 |
+
"error": "rate_limit",
|
113 |
+
"message": str(e),
|
114 |
+
"details": getattr(e, 'args', None)
|
115 |
+
}
|
116 |
+
except Exception as e:
|
117 |
+
print(f"Error in generate method: {str(e)}")
|
118 |
+
print(f"Error type: {type(e).__name__}")
|
119 |
+
print(f"Error details: {e.args}")
|
120 |
+
return {
|
121 |
+
"error": type(e).__name__,
|
122 |
+
"message": str(e),
|
123 |
+
"details": getattr(e, 'args', None)
|
124 |
+
}
|
125 |
+
|
126 |
+
def _generate_text(
|
127 |
+
self, prompt, system_prompt=None, temperature=0.5, max_tokens=4000, top_p=0.99, response_format=None
|
128 |
+
):
|
129 |
+
|
130 |
+
sys_prompt_arg = system_prompt if system_prompt else self.system_prompt
|
131 |
+
|
132 |
+
if self.enable_cache:
|
133 |
+
cache_key = sys_prompt_arg + prompt
|
134 |
+
cache_or_none = self._check_cache(cache_key)
|
135 |
+
if cache_or_none is not None:
|
136 |
+
return cache_or_none
|
137 |
+
|
138 |
+
if self.model_string in ['o1', 'o1-mini']: # only supports base response currently
|
139 |
+
# print(f"Using structured model: {self.model_string}")
|
140 |
+
response = self.client.beta.chat.completions.parse(
|
141 |
+
model=self.model_string,
|
142 |
+
messages=[
|
143 |
+
{"role": "user", "content": prompt},
|
144 |
+
],
|
145 |
+
max_completion_tokens=max_tokens
|
146 |
+
)
|
147 |
+
if response.choices[0].finishreason == "length":
|
148 |
+
response = "Token limit exceeded"
|
149 |
+
else:
|
150 |
+
response = response.choices[0].message.parsed
|
151 |
+
elif self.model_string in OPENAI_STRUCTURED_MODELS and response_format is not None:
|
152 |
+
# print(f"Using structured model: {self.model_string}")
|
153 |
+
response = self.client.beta.chat.completions.parse(
|
154 |
+
model=self.model_string,
|
155 |
+
messages=[
|
156 |
+
{"role": "system", "content": sys_prompt_arg},
|
157 |
+
{"role": "user", "content": prompt},
|
158 |
+
],
|
159 |
+
frequency_penalty=0,
|
160 |
+
presence_penalty=0,
|
161 |
+
stop=None,
|
162 |
+
temperature=temperature,
|
163 |
+
max_tokens=max_tokens,
|
164 |
+
top_p=top_p,
|
165 |
+
response_format=response_format
|
166 |
+
)
|
167 |
+
response = response.choices[0].message.parsed
|
168 |
+
else:
|
169 |
+
# print(f"Using non-structured model: {self.model_string}")
|
170 |
+
response = self.client.chat.completions.create(
|
171 |
+
model=self.model_string,
|
172 |
+
messages=[
|
173 |
+
{"role": "system", "content": sys_prompt_arg},
|
174 |
+
{"role": "user", "content": prompt},
|
175 |
+
],
|
176 |
+
frequency_penalty=0,
|
177 |
+
presence_penalty=0,
|
178 |
+
stop=None,
|
179 |
+
temperature=temperature,
|
180 |
+
max_tokens=max_tokens,
|
181 |
+
top_p=top_p,
|
182 |
+
)
|
183 |
+
response = response.choices[0].message.content
|
184 |
+
|
185 |
+
if self.enable_cache:
|
186 |
+
self._save_cache(cache_key, response)
|
187 |
+
return response
|
188 |
+
|
189 |
+
def __call__(self, prompt, **kwargs):
|
190 |
+
return self.generate(prompt, **kwargs)
|
191 |
+
|
192 |
+
def _format_content(self, content: List[Union[str, bytes]]) -> List[dict]:
|
193 |
+
formatted_content = []
|
194 |
+
for item in content:
|
195 |
+
if isinstance(item, bytes):
|
196 |
+
base64_image = base64.b64encode(item).decode('utf-8')
|
197 |
+
formatted_content.append({
|
198 |
+
"type": "image_url",
|
199 |
+
"image_url": {
|
200 |
+
"url": f"data:image/jpeg;base64,{base64_image}"
|
201 |
+
}
|
202 |
+
})
|
203 |
+
elif isinstance(item, str):
|
204 |
+
formatted_content.append({
|
205 |
+
"type": "text",
|
206 |
+
"text": item
|
207 |
+
})
|
208 |
+
else:
|
209 |
+
raise ValueError(f"Unsupported input type: {type(item)}")
|
210 |
+
return formatted_content
|
211 |
+
|
212 |
+
def _generate_multimodal(
|
213 |
+
self, content: List[Union[str, bytes]], system_prompt=None, temperature=0.5, max_tokens=4000, top_p=0.99, response_format=None
|
214 |
+
):
|
215 |
+
sys_prompt_arg = system_prompt if system_prompt else self.system_prompt
|
216 |
+
formatted_content = self._format_content(content)
|
217 |
+
|
218 |
+
if self.enable_cache:
|
219 |
+
cache_key = sys_prompt_arg + json.dumps(formatted_content)
|
220 |
+
cache_or_none = self._check_cache(cache_key)
|
221 |
+
if cache_or_none is not None:
|
222 |
+
# print(f"Cache hit for prompt: {cache_key[:200]}")
|
223 |
+
return cache_or_none
|
224 |
+
|
225 |
+
if self.model_string in ['o1', 'o1-mini']: # only supports base response currently
|
226 |
+
# print(f"Using structured model: {self.model_string}")
|
227 |
+
print(f'Max tokens: {max_tokens}')
|
228 |
+
response = self.client.chat.completions.create(
|
229 |
+
model=self.model_string,
|
230 |
+
messages=[
|
231 |
+
{"role": "user", "content": formatted_content},
|
232 |
+
],
|
233 |
+
max_completion_tokens=max_tokens
|
234 |
+
)
|
235 |
+
if response.choices[0].finish_reason == "length":
|
236 |
+
response_text = "Token limit exceeded"
|
237 |
+
else:
|
238 |
+
response_text = response.choices[0].message.content
|
239 |
+
elif self.model_string in OPENAI_STRUCTURED_MODELS and response_format is not None:
|
240 |
+
# print(f"Using structured model: {self.model_string}")
|
241 |
+
response = self.client.beta.chat.completions.parse(
|
242 |
+
model=self.model_string,
|
243 |
+
messages=[
|
244 |
+
{"role": "system", "content": sys_prompt_arg},
|
245 |
+
{"role": "user", "content": formatted_content},
|
246 |
+
],
|
247 |
+
temperature=temperature,
|
248 |
+
max_tokens=max_tokens,
|
249 |
+
top_p=top_p,
|
250 |
+
response_format=response_format
|
251 |
+
)
|
252 |
+
response_text = response.choices[0].message.parsed
|
253 |
+
else:
|
254 |
+
# print(f"Using non-structured model: {self.model_string}")
|
255 |
+
response = self.client.chat.completions.create(
|
256 |
+
model=self.model_string,
|
257 |
+
messages=[
|
258 |
+
{"role": "system", "content": sys_prompt_arg},
|
259 |
+
{"role": "user", "content": formatted_content},
|
260 |
+
],
|
261 |
+
temperature=temperature,
|
262 |
+
max_tokens=max_tokens,
|
263 |
+
top_p=top_p,
|
264 |
+
)
|
265 |
+
response_text = response.choices[0].message.content
|
266 |
+
|
267 |
+
if self.enable_cache:
|
268 |
+
self._save_cache(cache_key, response_text)
|
269 |
+
return response_text
|