codekingpro/portable-devtools
114k
1"""Wrapper around Konko AI's Completion API."""2 3import logging4import warnings5from typing import Any, Dict, List, Optional6 7from langchain_core.callbacks import (8 AsyncCallbackManagerForLLMRun,9 CallbackManagerForLLMRun,10)11from langchain_core.language_models.llms import LLM12from pydantic import ConfigDict, SecretStr, model_validator13 14from langchain_community.utils.openai import is_openai_v115 16logger = logging.getLogger(__name__)17 18 19class Konko(LLM):20 """Konko AI models.21 22 To use, you'll need an API key. This can be passed in as init param23 ``konko_api_key`` or set as environment variable ``KONKO_API_KEY``.24 25 Konko AI API reference: https://docs.konko.ai/reference/26 """27 28 base_url: str = "https://api.konko.ai/v1/completions"29 """Base inference API URL."""30 konko_api_key: SecretStr31 """Konko AI API key."""32 model: str33 """Model name. Available models listed here: 34 https://docs.konko.ai/reference/get_models35 """36 temperature: Optional[float] = None37 """Model temperature."""38 top_p: Optional[float] = None39 """Used to dynamically adjust the number of choices for each predicted token based 40 on the cumulative probabilities. A value of 1 will always yield the same 41 output. A temperature less than 1 favors more correctness and is appropriate 42 for question answering or summarization. A value greater than 1 introduces more 43 randomness in the output.44 """45 top_k: Optional[int] = None46 """Used to limit the number of choices for the next predicted word or token. It 47 specifies the maximum number of tokens to consider at each step, based on their 48 probability of occurrence. This technique helps to speed up the generation 49 process and can improve the quality of the generated text by focusing on the 50 most likely options.51 """52 max_tokens: Optional[int] = None53 """The maximum number of tokens to generate."""54 repetition_penalty: Optional[float] = None55 """A number that controls the diversity of generated text by reducing the 56 likelihood of repeated sequences. Higher values decrease repetition.57 """58 logprobs: Optional[int] = None59 """An integer that specifies how many top token log probabilities are included in 60 the response for each token generation step.61 """62 63 model_config = ConfigDict(64 extra="forbid",65 )66 67 @model_validator(mode="before")68 @classmethod69 def validate_environment(cls, values: Dict[str, Any]) -> Any:70 """Validate that python package exists in environment."""71 try:72 import konko73 74 except ImportError:75 raise ImportError(76 "Could not import konko python package. "77 "Please install it with `pip install konko`."78 )79 if not hasattr(konko, "_is_legacy_openai"):80 warnings.warn(81 "You are using an older version of the 'konko' package. "82 "Please consider upgrading to access new features"83 "including the completion endpoint."84 )85 return values86 87 def construct_payload(88 self,89 prompt: str,90 stop: Optional[List[str]] = None,91 **kwargs: Any,92 ) -> Dict[str, Any]:93 stop_to_use = stop[0] if stop and len(stop) == 1 else stop94 payload: Dict[str, Any] = {95 **self.default_params,96 "prompt": prompt,97 "stop": stop_to_use,98 **kwargs,99 }100 return {k: v for k, v in payload.items() if v is not None}101 102 @property103 def _llm_type(self) -> str:104 """Return type of model."""105 return "konko"106 107 @staticmethod108 def get_user_agent() -> str:109 from langchain_community import __version__110 111 return f"langchain/{__version__}"112 113 @property114 def default_params(self) -> Dict[str, Any]:115 return {116 "model": self.model,117 "temperature": self.temperature,118 "top_p": self.top_p,119 "top_k": self.top_k,120 "max_tokens": self.max_tokens,121 "repetition_penalty": self.repetition_penalty,122 }123 124 def _call(125 self,126 prompt: str,127 stop: Optional[List[str]] = None,128 run_manager: Optional[CallbackManagerForLLMRun] = None,129 **kwargs: Any,130 ) -> str:131 """Call out to Konko's text generation endpoint.132 133 Args:134 prompt: The prompt to pass into the model.135 136 Returns:137 The string generated by the model..138 """139 import konko140 141 payload = self.construct_payload(prompt, stop, **kwargs)142 143 try:144 if is_openai_v1():145 response = konko.completions.create(**payload)146 else:147 response = konko.Completion.create(**payload)148 149 except AttributeError:150 raise ValueError(151 "`konko` has no `Completion` attribute, this is likely "152 "due to an old version of the konko package. Try upgrading it "153 "with `pip install --upgrade konko`."154 )155 156 if is_openai_v1():157 output = response.choices[0].text158 else:159 output = response["choices"][0]["text"]160 161 return output162 163 async def _acall(164 self,165 prompt: str,166 stop: Optional[List[str]] = None,167 run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,168 **kwargs: Any,169 ) -> str:170 """Asynchronously call out to Konko's text generation endpoint.171 172 Args:173 prompt: The prompt to pass into the model.174 175 Returns:176 The string generated by the model.177 """178 import konko179 180 payload = self.construct_payload(prompt, stop, **kwargs)181 182 try:183 if is_openai_v1():184 client = konko.AsyncKonko()185 response = await client.completions.create(**payload)186 else:187 response = await konko.Completion.acreate(**payload)188 189 except AttributeError:190 raise ValueError(191 "`konko` has no `Completion` attribute, this is likely "192 "due to an old version of the konko package. Try upgrading it "193 "with `pip install --upgrade konko`."194 )195 196 if is_openai_v1():197 output = response.choices[0].text198 else:199 output = response["choices"][0]["text"]200 201 return output202 