Spaces:
Build error
Build error
fix openai api base bug
Browse files- agentverse/llms/openai.py +24 -39
- app.py +0 -1
agentverse/llms/openai.py
CHANGED
@@ -5,7 +5,7 @@ import os
|
|
5 |
import numpy as np
|
6 |
from aiohttp import ClientSession
|
7 |
from typing import Dict, List, Optional, Union
|
8 |
-
from tenacity import retry, stop_after_attempt, wait_exponential
|
9 |
|
10 |
from pydantic import BaseModel, Field
|
11 |
|
@@ -37,15 +37,20 @@ else:
|
|
37 |
openai.api_version = "2023-05-15"
|
38 |
is_openai_available = True
|
39 |
else:
|
40 |
-
logging.warning(
|
41 |
-
"OpenAI API key is not set. Please set the environment variable OPENAI_API_KEY"
|
42 |
-
)
|
43 |
is_openai_available = False
|
44 |
|
45 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
class OpenAIChatArgs(BaseModelArgs):
|
47 |
model: str = Field(default="gpt-3.5-turbo")
|
48 |
-
deployment_id: str = Field(default=
|
49 |
max_tokens: int = Field(default=2048)
|
50 |
temperature: float = Field(default=1.0)
|
51 |
top_p: int = Field(default=1)
|
@@ -117,6 +122,7 @@ class OpenAIChat(BaseChatModel):
|
|
117 |
stop=stop_after_attempt(20),
|
118 |
wait=wait_exponential(multiplier=1, min=4, max=10),
|
119 |
reraise=True,
|
|
|
120 |
)
|
121 |
def generate_response(
|
122 |
self,
|
@@ -139,13 +145,9 @@ class OpenAIChat(BaseChatModel):
|
|
139 |
self.collect_metrics(response)
|
140 |
return LLMResult(
|
141 |
content=response["choices"][0]["message"].get("content", ""),
|
142 |
-
function_name=response["choices"][0]["message"][
|
143 |
-
"function_call"
|
144 |
-
]["name"],
|
145 |
function_arguments=ast.literal_eval(
|
146 |
-
response["choices"][0]["message"]["function_call"][
|
147 |
-
"arguments"
|
148 |
-
]
|
149 |
),
|
150 |
send_tokens=response["usage"]["prompt_tokens"],
|
151 |
recv_tokens=response["usage"]["completion_tokens"],
|
@@ -179,6 +181,7 @@ class OpenAIChat(BaseChatModel):
|
|
179 |
stop=stop_after_attempt(20),
|
180 |
wait=wait_exponential(multiplier=1, min=4, max=10),
|
181 |
reraise=True,
|
|
|
182 |
)
|
183 |
async def agenerate_response(
|
184 |
self,
|
@@ -200,9 +203,7 @@ class OpenAIChat(BaseChatModel):
|
|
200 |
**self.args.dict(),
|
201 |
)
|
202 |
if response["choices"][0]["message"].get("function_call") is not None:
|
203 |
-
function_name = response["choices"][0]["message"]["function_call"][
|
204 |
-
"name"
|
205 |
-
]
|
206 |
valid_function = False
|
207 |
if function_name.startswith("function."):
|
208 |
function_name = function_name.replace("function.", "")
|
@@ -220,27 +221,15 @@ class OpenAIChat(BaseChatModel):
|
|
220 |
f"The returned function name {function_name} is not in the list of valid functions."
|
221 |
)
|
222 |
try:
|
223 |
-
arguments = ast.literal_eval(
|
224 |
-
response["choices"][0]["message"]["function_call"][
|
225 |
-
"arguments"
|
226 |
-
]
|
227 |
-
)
|
228 |
except:
|
229 |
try:
|
230 |
arguments = ast.literal_eval(
|
231 |
-
JsonRepair(
|
232 |
-
response["choices"][0]["message"]["function_call"][
|
233 |
-
"arguments"
|
234 |
-
]
|
235 |
-
).repair()
|
236 |
)
|
237 |
except:
|
238 |
-
logger.warn(
|
239 |
-
|
240 |
-
)
|
241 |
-
raise ValueError(
|
242 |
-
"The returned argument in function call is not valid json."
|
243 |
-
)
|
244 |
self.collect_metrics(response)
|
245 |
return LLMResult(
|
246 |
function_name=function_name,
|
@@ -276,9 +265,7 @@ class OpenAIChat(BaseChatModel):
|
|
276 |
except (OpenAIError, KeyboardInterrupt, json.decoder.JSONDecodeError) as error:
|
277 |
raise
|
278 |
|
279 |
-
def construct_messages(
|
280 |
-
self, prepend_prompt: str, history: List[dict], append_prompt: str
|
281 |
-
):
|
282 |
messages = []
|
283 |
if prepend_prompt != "":
|
284 |
messages.append({"role": "system", "content": prepend_prompt})
|
@@ -332,13 +319,11 @@ def get_embedding(text: str, attempts=3) -> np.array:
|
|
332 |
try:
|
333 |
text = text.replace("\n", " ")
|
334 |
if openai.api_type == "azure":
|
335 |
-
embedding = openai.Embedding.create(
|
336 |
-
|
337 |
-
|
338 |
else:
|
339 |
-
embedding = openai.Embedding.create(
|
340 |
-
input=[text], model="text-embedding-ada-002"
|
341 |
-
)["data"][0]["embedding"]
|
342 |
return tuple(embedding)
|
343 |
except Exception as e:
|
344 |
attempts += 1
|
|
|
5 |
import numpy as np
|
6 |
from aiohttp import ClientSession
|
7 |
from typing import Dict, List, Optional, Union
|
8 |
+
from tenacity import retry, stop_after_attempt, wait_exponential, RetryCallState
|
9 |
|
10 |
from pydantic import BaseModel, Field
|
11 |
|
|
|
37 |
openai.api_version = "2023-05-15"
|
38 |
is_openai_available = True
|
39 |
else:
|
40 |
+
logging.warning("OpenAI API key is not set. Please set the environment variable OPENAI_API_KEY")
|
|
|
|
|
41 |
is_openai_available = False
|
42 |
|
43 |
|
44 |
+
def log_retry(retry_state: RetryCallState):
|
45 |
+
exception = retry_state.outcome.exception()
|
46 |
+
logger.warn(
|
47 |
+
f"Retrying {retry_state.fn}\nAttempt: {retry_state.attempt_number}\nException: {exception.__class__.__name__} {exception}",
|
48 |
+
)
|
49 |
+
|
50 |
+
|
51 |
class OpenAIChatArgs(BaseModelArgs):
|
52 |
model: str = Field(default="gpt-3.5-turbo")
|
53 |
+
deployment_id: Optional[str] = Field(default=None)
|
54 |
max_tokens: int = Field(default=2048)
|
55 |
temperature: float = Field(default=1.0)
|
56 |
top_p: int = Field(default=1)
|
|
|
122 |
stop=stop_after_attempt(20),
|
123 |
wait=wait_exponential(multiplier=1, min=4, max=10),
|
124 |
reraise=True,
|
125 |
+
before_sleep=log_retry
|
126 |
)
|
127 |
def generate_response(
|
128 |
self,
|
|
|
145 |
self.collect_metrics(response)
|
146 |
return LLMResult(
|
147 |
content=response["choices"][0]["message"].get("content", ""),
|
148 |
+
function_name=response["choices"][0]["message"]["function_call"]["name"],
|
|
|
|
|
149 |
function_arguments=ast.literal_eval(
|
150 |
+
response["choices"][0]["message"]["function_call"]["arguments"]
|
|
|
|
|
151 |
),
|
152 |
send_tokens=response["usage"]["prompt_tokens"],
|
153 |
recv_tokens=response["usage"]["completion_tokens"],
|
|
|
181 |
stop=stop_after_attempt(20),
|
182 |
wait=wait_exponential(multiplier=1, min=4, max=10),
|
183 |
reraise=True,
|
184 |
+
before_sleep=log_retry,
|
185 |
)
|
186 |
async def agenerate_response(
|
187 |
self,
|
|
|
203 |
**self.args.dict(),
|
204 |
)
|
205 |
if response["choices"][0]["message"].get("function_call") is not None:
|
206 |
+
function_name = response["choices"][0]["message"]["function_call"]["name"]
|
|
|
|
|
207 |
valid_function = False
|
208 |
if function_name.startswith("function."):
|
209 |
function_name = function_name.replace("function.", "")
|
|
|
221 |
f"The returned function name {function_name} is not in the list of valid functions."
|
222 |
)
|
223 |
try:
|
224 |
+
arguments = ast.literal_eval(response["choices"][0]["message"]["function_call"]["arguments"])
|
|
|
|
|
|
|
|
|
225 |
except:
|
226 |
try:
|
227 |
arguments = ast.literal_eval(
|
228 |
+
JsonRepair(response["choices"][0]["message"]["function_call"]["arguments"]).repair()
|
|
|
|
|
|
|
|
|
229 |
)
|
230 |
except:
|
231 |
+
logger.warn("The returned argument in function call is not valid json. Retrying...")
|
232 |
+
raise ValueError("The returned argument in function call is not valid json.")
|
|
|
|
|
|
|
|
|
233 |
self.collect_metrics(response)
|
234 |
return LLMResult(
|
235 |
function_name=function_name,
|
|
|
265 |
except (OpenAIError, KeyboardInterrupt, json.decoder.JSONDecodeError) as error:
|
266 |
raise
|
267 |
|
268 |
+
def construct_messages(self, prepend_prompt: str, history: List[dict], append_prompt: str):
|
|
|
|
|
269 |
messages = []
|
270 |
if prepend_prompt != "":
|
271 |
messages.append({"role": "system", "content": prepend_prompt})
|
|
|
319 |
try:
|
320 |
text = text.replace("\n", " ")
|
321 |
if openai.api_type == "azure":
|
322 |
+
embedding = openai.Embedding.create(input=[text], deployment_id="text-embedding-ada-002")["data"][0][
|
323 |
+
"embedding"
|
324 |
+
]
|
325 |
else:
|
326 |
+
embedding = openai.Embedding.create(input=[text], model="text-embedding-ada-002")["data"][0]["embedding"]
|
|
|
|
|
327 |
return tuple(embedding)
|
328 |
except Exception as e:
|
329 |
attempts += 1
|
app.py
CHANGED
@@ -335,7 +335,6 @@ class GUI:
|
|
335 |
"""
|
336 |
|
337 |
# data = self.backend.next_data()
|
338 |
-
|
339 |
return_message = self.backend.next()
|
340 |
|
341 |
data = self.return_format(return_message)
|
|
|
335 |
"""
|
336 |
|
337 |
# data = self.backend.next_data()
|
|
|
338 |
return_message = self.backend.next()
|
339 |
|
340 |
data = self.return_format(return_message)
|