xinyin025/LangBot
1
0
Fork 0
mirror of https://github.com/langbot-app/LangBot.git synced 2026-07-20 19:36:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: 不再预先计算前文token数而是在报错时提醒用户重置

Browse Source
This commit is contained in:
RockChinQ
2024-03-12 16:04:11 +08:00
parent a398c6f311
commit 1d963d0f0c
16 changed files with 25 additions and 144 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pkg/provider/requester/__init__.py → pkg/provider/modelmgr/__init__.py
View File
pkg/provider/requester/api.py → pkg/provider/modelmgr/api.py
+14
View File
@@ -7,9 +7,23 @@ from ...core import app
from ...core import entities as core_entities
from .. import entities as llm_entities
preregistered_requesters: list[typing.Type[LLMAPIRequester]] = []
def requester_class(name: str):
def decorator(cls: typing.Type[LLMAPIRequester]) -> typing.Type[LLMAPIRequester]:
cls.name = name
preregistered_requesters.append(cls)
return cls
return decorator
class LLMAPIRequester(metaclass=abc.ABCMeta):
"""LLM API请求器
"""
name: str = None
ap: app.Application
pkg/provider/requester/apis/__init__.py → pkg/provider/modelmgr/apis/__init__.py
View File
pkg/provider/requester/apis/chatcmpl.py → pkg/provider/modelmgr/apis/chatcmpl.py
+5 -1
View File
@@ -17,6 +17,7 @@ from ... import entities as llm_entities
from ...tools import entities as tools_entities
@api.requester_class("openai-chat-completion")
class OpenAIChatCompletion(api.LLMAPIRequester):
"""OpenAI ChatCompletion API 请求器"""
@@ -133,7 +134,10 @@ class OpenAIChatCompletion(api.LLMAPIRequester):
except asyncio.TimeoutError:
raise errors.RequesterError('请求超时')
except openai.BadRequestError as e:
raise errors.RequesterError(f'请求错误: {e.message}')
if 'context_length_exceeded' in e.message:
raise errors.RequesterError(f'上文过长，请重置会话: {e.message}')
else:
raise errors.RequesterError(f'请求参数错误: {e.message}')
except openai.AuthenticationError as e:
raise errors.RequesterError(f'无效的 api-key: {e.message}')
except openai.NotFoundError as e:
pkg/provider/requester/entities.py → pkg/provider/modelmgr/entities.py
+1 -5
View File
@@ -5,7 +5,7 @@ import typing
import pydantic
from . import api
from . import token, tokenizer
from . import token
class LLMModelInfo(pydantic.BaseModel):
@@ -19,11 +19,7 @@ class LLMModelInfo(pydantic.BaseModel):
requester: api.LLMAPIRequester
tokenizer: 'tokenizer.LLMTokenizer'
tool_call_supported: typing.Optional[bool] = False
max_tokens: typing.Optional[int] = 2048
class Config:
arbitrary_types_allowed = True
pkg/provider/requester/errors.py → pkg/provider/modelmgr/errors.py
View File
pkg/provider/requester/modelmgr.py → pkg/provider/modelmgr/modelmgr.py
+2 -51
View File
@@ -3,9 +3,8 @@ from __future__ import annotations
from . import entities
from ...core import app
from .apis import chatcmpl
from . import token
from .tokenizers import tiktoken
from .apis import chatcmpl
class ModelManager:
@@ -30,9 +29,7 @@ class ModelManager:
async def initialize(self):
openai_chat_completion = chatcmpl.OpenAIChatCompletion(self.ap)
await openai_chat_completion.initialize()
openai_token_mgr = token.TokenManager(self.ap, list(self.ap.provider_cfg.data['openai-config']['api-keys']))
tiktoken_tokenizer = tiktoken.Tiktoken(self.ap)
openai_token_mgr = token.TokenManager("openai", list(self.ap.provider_cfg.data['openai-config']['api-keys']))
model_list = [
entities.LLMModelInfo(
@@ -40,48 +37,36 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=4096
),
entities.LLMModelInfo(
name="gpt-3.5-turbo-1106",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=16385
),
entities.LLMModelInfo(
name="gpt-3.5-turbo-16k",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=16385
),
entities.LLMModelInfo(
name="gpt-3.5-turbo-0613",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=4096
),
entities.LLMModelInfo(
name="gpt-3.5-turbo-16k-0613",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=16385
),
entities.LLMModelInfo(
name="gpt-3.5-turbo-0301",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=4096
)
]
@@ -93,64 +78,48 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=128000
),
entities.LLMModelInfo(
name="gpt-4-turbo-preview",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=128000
),
entities.LLMModelInfo(
name="gpt-4-1106-preview",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=128000
),
entities.LLMModelInfo(
name="gpt-4-vision-preview",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=128000
),
entities.LLMModelInfo(
name="gpt-4",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=8192
),
entities.LLMModelInfo(
name="gpt-4-0613",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=8192
),
entities.LLMModelInfo(
name="gpt-4-32k",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=32768
),
entities.LLMModelInfo(
name="gpt-4-32k-0613",
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=True,
tokenizer=tiktoken_tokenizer,
max_tokens=32768
)
]
@@ -163,8 +132,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=8192
),
entities.LLMModelInfo(
name="OneAPI/chatglm_pro",
@@ -172,8 +139,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=128000
),
entities.LLMModelInfo(
name="OneAPI/chatglm_std",
@@ -181,8 +146,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=128000
),
entities.LLMModelInfo(
name="OneAPI/chatglm_lite",
@@ -190,8 +153,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=128000
),
entities.LLMModelInfo(
name="OneAPI/qwen-v1",
@@ -199,8 +160,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=6000
),
entities.LLMModelInfo(
name="OneAPI/qwen-plus-v1",
@@ -208,8 +167,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=30000
),
entities.LLMModelInfo(
name="OneAPI/ERNIE-Bot",
@@ -217,8 +174,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=2000
),
entities.LLMModelInfo(
name="OneAPI/ERNIE-Bot-turbo",
@@ -226,8 +181,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=7000
),
entities.LLMModelInfo(
name="OneAPI/gemini-pro",
@@ -235,8 +188,6 @@ class ModelManager:
token_mgr=openai_token_mgr,
requester=openai_chat_completion,
tool_call_supported=False,
tokenizer=tiktoken_tokenizer,
max_tokens=30720
),
]
pkg/provider/requester/token.py → pkg/provider/modelmgr/token.py
View File
pkg/provider/requester/tokenizer.py
-30
View File
@@ -1,30 +0,0 @@
from __future__ import annotations
import abc
import typing
from ...core import app
from .. import entities as llm_entities
from . import entities
class LLMTokenizer(metaclass=abc.ABCMeta):
"""LLM分词器抽象类"""
ap: app.Application
def __init__(self, ap: app.Application):
self.ap = ap
async def initialize(self):
"""初始化分词器
"""
pass
@abc.abstractmethod
async def count_token(
self,
messages: list[llm_entities.Message],
model: entities.LLMModelInfo
) -> int:
pass
pkg/provider/requester/tokenizers/__init__.py
View File
pkg/provider/requester/tokenizers/tiktoken.py
-30
View File
@@ -1,30 +0,0 @@
from __future__ import annotations
import tiktoken
from .. import tokenizer
from ... import entities as llm_entities
from .. import entities
class Tiktoken(tokenizer.LLMTokenizer):
"""TikToken分词器
"""
async def count_token(
self,
messages: list[llm_entities.Message],
model: entities.LLMModelInfo
) -> int:
try:
encoding = tiktoken.encoding_for_model(model.name)
except KeyError:
# print("Warning: model not found. Using cl100k_base encoding.")
encoding = tiktoken.get_encoding("cl100k_base")
num_tokens = 0
for message in messages:
num_tokens += len(encoding.encode(message.role))
num_tokens += len(encoding.encode(message.content if message.content is not None else ''))
num_tokens += 3 # every reply is primed with <|start|>assistant<|message|>
return num_tokens