1 line
17 KiB
JSON
1 line
17 KiB
JSON
{"repo": "aiahmedshrf/mybisheng", "files": {"src/backend/bisheng_langchain/chat_models/sensetime.py": {"79eef2140735143dbe71f7b31494775caa7ddb43": {"t": 1785768275.1909099, "n": 16949, "h": "29a2a43980462f5e", "c": "from __future__ import annotations\n\nimport ast\nimport json\nimport logging\nimport time\nfrom typing import Any, Dict, List, Mapping, Optional, Tuple, Union\n\nimport jwt\nfrom pydantic import ConfigDict, model_validator, Field\n\nfrom bisheng_langchain.utils.requests import Requests\nfrom langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun\nfrom langchain.chat_models.base import BaseChatModel\nfrom langchain.schema import ChatGeneration, ChatResult\nfrom langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,\n HumanMessage, SystemMessage)\nfrom langchain.utils import get_from_dict_or_env\nfrom tenacity import (before_sleep_log, retry, retry_if_exception_type, stop_after_attempt,\n wait_exponential)\n\n# if TYPE_CHECKING:\n# import jwt\n\nlogger = logging.getLogger(__name__)\n\n\ndef _import_pyjwt() -> Any:\n try:\n import jwt\n except ImportError:\n raise ValueError('Could not import jwt python package. '\n 'This is needed in order to calculate get_token_ids. '\n 'Please install it with `pip install PyJWT`.')\n return jwt\n\n\ndef encode_jwt_token(ak, sk):\n headers = {'alg': 'HS256', 'typ': 'JWT'}\n payload = {\n 'iss': ak,\n 'exp': int(time.time()) + 18000, # 填写您期望的有效时间，此处示例代表当前时间+300分钟\n 'nbf': int(time.time()) - 500 # 填写您期望的生效时间，此处示例代表当前时间-500秒\n }\n token = jwt.encode(payload, sk, headers=headers)\n return token\n\n\ndef _create_retry_decorator(llm):\n\n min_seconds = 1\n max_seconds = 20\n # Wait 2^x * 1 second between each retry starting with\n # 4 seconds, then up to 10 seconds, then 10 seconds afterwards\n return retry(\n reraise=True,\n stop=stop_after_attempt(llm.max_retries),\n wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),\n retry=(retry_if_exception_type(Exception)),\n before_sleep=before_sleep_log(logger, logging.WARNING),\n )\n\n\ndef _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:\n role = _dict['role']\n if role == 'user':\n return HumanMessage(content=_dict['content'])\n elif role == 'assistant':\n content = _dict['content'] or '' # OpenAI returns None for tool invocations\n\n if _dict.get('function_call'):\n additional_kwargs = {'function_call': dict(_dict['function_call'])}\n else:\n additional_kwargs = {}\n return AIMessage(content=content, additional_kwargs=additional_kwargs)\n elif role == 'system':\n return SystemMessage(content=_dict['content'])\n elif role == 'function':\n return FunctionMessage(content=_dict['content'], name=_dict['name'])\n else:\n return ChatMessage(content=_dict['content'], role=role)\n\n\ndef _convert_message_to_dict(message: BaseMessage) -> dict:\n if isinstance(message, ChatMessage):\n message_dict = {'role': message.role, 'content': message.content}\n elif isinstance(message, HumanMessage):\n message_dict = {'role': 'user', 'content': message.content}\n elif isinstance(message, AIMessage):\n message_dict = {'role': 'assistant', 'content': message.content}\n elif isinstance(message, SystemMessage):\n message_dict = {'role': 'system', 'content': message.content}\n # raise ValueError(f\"not support system role {message}\")\n\n elif isinstance(message, FunctionMessage):\n raise ValueError(f'not support funciton {message}')\n else:\n raise ValueError(f'Got unknown type {message}')\n\n # if \"name\" in message.additional_kwargs:\n # message_dict[\"name\"] = message.additional_kwargs[\"name\"]\n return message_dict\n\n\ndef _convert_message_to_dict2(message: BaseMessage) -> List[dict]:\n if isinstance(message, ChatMessage):\n message_dict = {'role': message.role, 'content': message.content}\n elif isinstance(message, HumanMessage):\n message_dict = {'role': 'user', 'content': message.content}\n elif isinstance(message, AIMessage):\n message_dict = {'role': 'assistant', 'content': message.content}\n elif isinstance(message, SystemMessage):\n raise ValueError(f'not support system role {message}')\n\n elif isinstance(message, FunctionMessage):\n raise ValueError(f'not support funciton {message}')\n else:\n raise ValueError(f'Got unknown type {message}')\n\n return [message_dict]\n\n\nurl = 'https://api.sensenova.cn/v1/llm/chat-completions'\n\n\nclass SenseChat(BaseChatModel):\n\n client: Optional[Any] #: :meta private:\n model_name: str = Field(default='SenseChat', alias='model')\n \"\"\"Model name to use.\"\"\"\n temperature: float = 0.8\n top_p: float = 0.7\n \"\"\"What sampling temperature to use.\"\"\"\n model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)\n \"\"\"Holds any model parameters valid for `create` call not explicitly specified.\"\"\"\n access_key_id: Optional[str] = None\n secret_access_key: Optional[str] = None\n\n repetition_penalty: float = 1.05\n request_timeout: Optional[Union[float, Tuple[float, float]]] = None\n \"\"\"Timeout for requests to OpenAI completion API. Default is 600 seconds.\"\"\"\n max_retries: Optional[int] = 6\n \"\"\"Maximum number of retries to make when generating.\"\"\"\n streaming: Optional[bool] = False\n \"\"\"Whether to stream the results or not.\"\"\"\n n: Optional[int] = 1\n \"\"\"Number of chat completions to generate for each prompt.\"\"\"\n max_tokens: Optional[int] = 1024\n \"\"\"Maximum number of tokens to generate.\"\"\"\n tiktoken_model_name: Optional[str] = None\n \"\"\"The model name to pass to tiktoken when using this class.\n Tiktoken is used to count the number of tokens in documents to constrain\n them to be under a certain limit. By default, when set to None, this will\n be the same as the embedding model name. However, there are some cases\n where you may want to use this Embedding class with a model name not\n supported by tiktoken. This can include when using Azure embeddings or\n when using one of the many model providers that expose an OpenAI-like\n API but with different models. In those cases, in order to avoid erroring\n when tiktoken is called, you can specify a model name to use here.\"\"\"\n verbose: Optional[bool] = False\n model_config = ConfigDict(validate_by_name=True)\n\n @model_validator(mode='before')\n @classmethod\n def validate_environment(cls, values: Dict) -> Dict:\n \"\"\"Validate that api key and python package exists in environment.\"\"\"\n\n _import_pyjwt()\n\n values['access_key_id'] = get_from_dict_or_env(values, 'access_key_id', 'ACCESS_KEY_ID')\n values['secret_access_key'] = get_from_dict_or_env(values, 'secret_access_key',\n 'SECRET_ACCESS_KEY')\n token = encode_jwt_token(values['access_key_id'], values['secret_access_key'])\n if isinstance(token, bytes):\n token = token.decode('utf-8')\n\n try:\n header = {\n 'Authorization': 'Bearer {}'.format(token),\n 'Content-Type': 'application/json'\n }\n\n values['client'] = Requests(headers=header, )\n except AttributeError:\n raise ValueError('Try upgrading it with `pip install --upgrade requests`.')\n return values\n\n @property\n def _default_params(self) -> Dict[str, Any]:\n \"\"\"Get the default parameters for calling ZhipuAI API.\"\"\"\n return {\n 'model': self.model_name,\n 'temperature': self.temperature,\n 'top_p': self.top_p,\n 'max_tokens': self.max_tokens,\n **self.model_kwargs,\n }\n\n def completion_with_retry(self, **kwargs: Any) -> Any:\n retry_decorator = _create_retry_decorator(self)\n\n @retry_decorator\n def _completion_with_retry(**kwargs: Any) -> Any:\n messages = kwargs.get('messages')\n temperature = kwargs.get('temperature')\n top_p = kwargs.get('top_p')\n # messages\n params = {\n 'messages': messages,\n 'model': self.model_name,\n 'top_p': top_p,\n 'temperature': temperature,\n 'repetition_penalty': self.repetition_penalty,\n 'n': self.n,\n 'max_new_tokens': self.max_tokens,\n 'stream': False # self.streaming\n }\n\n token = encode_jwt_token(self.access_key_id, self.secret_access_key)\n if isinstance(token, bytes):\n token = token.decode('utf-8')\n self.client.headers.update({'Authorization': 'Bearer {}'.format(token)})\n\n response = self.client.post(url=url, json=params).json()\n return response\n\n rsp_dict = _completion_with_retry(**kwargs)\n if 'error' in rsp_dict:\n logger.error(f'sensechat_error resp={rsp_dict}')\n message = rsp_dict['error']['message']\n raise Exception(message)\n else:\n # return rsp_dict['data'], rsp_dict.get('usage', '')\n return rsp_dict, rsp_dict.get('usage', '')\n\n async def acompletion_with_retry(self, **kwargs: Any) -> Any:\n \"\"\"Use tenacity to retry the async completion call.\"\"\"\n retry_decorator = _create_retry_decorator(self)\n\n if self.streaming:\n self.client.headers.update({'Accept': 'text/event-stream'})\n else:\n self.client.headers.pop('Accept', '')\n\n @retry_decorator\n async def _acompletion_with_retry(**kwargs: Any) -> Any:\n messages = kwargs.pop('messages', '')\n\n inp = {\n 'messages': messages,\n 'model': self.model_name,\n 'top_p': self.top_p,\n 'temperature': self.temperature,\n 'repetition_penalty': self.repetition_penalty,\n 'n': self.n,\n 'max_new_tokens': self.max_tokens,\n 'stream': True\n }\n\n # Use OpenAI's async api https://github.com/openai/openai-python#async-api\n async with self.client.apost(url=url, json=inp) as response:\n\n async for line in response.content.iter_any():\n\n if b'\\n' in line:\n for txt_ in line.split(b'\\n'):\n yield txt_.decode('utf-8').strip()\n else:\n yield line.decode('utf-8').strip()\n\n async for response in _acompletion_with_retry(**kwargs):\n is_error = False\n if response:\n if response.startswith('event:error'):\n is_error = True\n elif response.startswith('data:'):\n yield (is_error, response[len('data:'):])\n if is_error:\n break\n elif response.startswith('{'):\n yield (is_error, response)\n else:\n continue\n\n def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:\n overall_token_usage: dict = {}\n for output in llm_outputs:\n if output is None:\n # Happens in streaming\n continue\n token_usage = output['token_usage']\n for k, v in token_usage.items():\n if k in overall_token_usage:\n overall_token_usage[k] += v\n else:\n overall_token_usage[k] = v\n return {'token_usage': overall_token_usage, 'model_name': self.model_name}\n\n def _generate(\n self,\n messages: List[BaseMessage],\n stop: Optional[List[str]] = None,\n run_manager: Optional[CallbackManagerForLLMRun] = None,\n **kwargs: Any,\n ) -> ChatResult:\n\n message_dicts, params = self._create_message_dicts(messages, stop)\n params = {**params, **kwargs}\n response, usage = self.completion_with_retry(messages=message_dicts, **params)\n\n return self._create_chat_result(response)\n\n async def _agenerate(\n self,\n messages: List[BaseMessage],\n stop: Optional[List[str]] = None,\n run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,\n **kwargs: Any,\n ) -> ChatResult:\n message_dicts, params = self._create_message_dicts(messages, stop)\n params = {**params, **kwargs}\n if self.streaming:\n\n inner_completion = ''\n role = 'user'\n params['stream'] = True\n function_call: Optional[dict] = None\n async for is_error, stream_resp in self.acompletion_with_retry(messages=message_dicts,\n **params):\n if str(stream_resp).startswith('[DONE]'):\n continue\n output = json.loads(stream_resp)\n if is_error:\n logger.error(stream_resp)\n raise ValueError(stream_resp)\n if 'data' in output:\n output = output['data']\n\n choices = None\n if 'choices' in output:\n choices = output.get('choices')\n\n if choices:\n for choice in choices:\n token = choice['delta']\n\n inner_completion += token or ''\n _function_call = ''\n if run_manager:\n await run_manager.on_llm_new_token(token)\n if _function_call:\n if function_call is None:\n function_call = _function_call\n else:\n function_call['arguments'] += _function_call['arguments']\n message = _convert_dict_to_message({\n 'content': inner_completion,\n 'role': role,\n 'function_call': function_call,\n })\n return ChatResult(generations=[ChatGeneration(message=message)])\n else:\n return self._generate(messages, stop, run_manager, **kwargs)\n\n def _create_message_dicts(\n self, messages: List[BaseMessage],\n stop: Optional[List[str]]) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:\n params = dict(self._client_params)\n if stop is not None:\n if 'stop' in params:\n raise ValueError('`stop` found in both the input and default params.')\n params['stop'] = stop\n\n system_content = ''\n message_dicts = []\n for m in messages:\n if m.type == 'system':\n system_content += m.content\n continue\n message_dicts.extend(_convert_message_to_dict2(m))\n\n if system_content:\n message_dicts[-1]['content'] = system_content + message_dicts[-1]['content']\n\n return message_dicts, params\n\n def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:\n generations = []\n\n def _norm_text(text):\n if text[0] == '\"' and text[-1] == '\"':\n out = ast.literal_eval(text)\n else:\n out = text\n return out\n\n for res in response['data']['choices']:\n res['content'] = _norm_text(res['message'])\n res['role'] = 'user'\n message = _convert_dict_to_message(res)\n gen = ChatGeneration(message=message)\n generations.append(gen)\n\n llm_output = {'token_usage': response['data']['usage'], 'model_name': self.model_name}\n return ChatResult(generations=generations, llm_output=llm_output)\n\n @property\n def _identifying_params(self) -> Mapping[str, Any]:\n \"\"\"Get the identifying parameters.\"\"\"\n return {**{'model_name': self.model_name}, **self._default_params}\n\n @property\n def _client_params(self) -> Mapping[str, Any]:\n \"\"\"Get the parameters used for the openai client.\"\"\"\n zhipu_creds: Dict[str, Any] = {\n 'access_key_id': self.access_key_id,\n 'secret_access_key': self.secret_access_key,\n 'model': self.model_name,\n }\n return {**zhipu_creds, **self._default_params}\n\n def _get_invocation_params(self,\n stop: Optional[List[str]] = None,\n **kwargs: Any) -> Dict[str, Any]:\n \"\"\"Get the parameters used to invoke the model FOR THE CALLBACKS.\"\"\"\n return {\n **super()._get_invocation_params(stop=stop, **kwargs),\n **self._default_params,\n 'model': self.model_name,\n 'function': kwargs.get('functions'),\n }\n\n @property\n def _llm_type(self) -> str:\n \"\"\"Return type of chat model.\"\"\"\n return 'sense-chat'\n"}}}} |