langchain/langchain/llms/writer.py
leo-gan fd69cc7e42
Removed duplicate BaseModel dependencies (#2471)
Removed duplicate BaseModel dependencies in class inheritances.
Also, sorted imports by `isort`.
2023-04-06 12:45:16 -07:00

156 lines
4.9 KiB
Python

"""Wrapper around Writer APIs."""
from typing import Any, Dict, List, Mapping, Optional
import requests
from pydantic import Extra, root_validator
from langchain.llms.base import LLM
from langchain.llms.utils import enforce_stop_tokens
from langchain.utils import get_from_dict_or_env
class Writer(LLM):
"""Wrapper around Writer large language models.
To use, you should have the environment variable ``WRITER_API_KEY``
set with your API key.
Example:
.. code-block:: python
from langchain import Writer
writer = Writer(model_id="palmyra-base")
"""
model_id: str = "palmyra-base"
"""Model name to use."""
tokens_to_generate: int = 24
"""Max number of tokens to generate."""
logprobs: bool = False
"""Whether to return log probabilities."""
temperature: float = 1.0
"""What sampling temperature to use."""
length: int = 256
"""The maximum number of tokens to generate in the completion."""
top_p: float = 1.0
"""Total probability mass of tokens to consider at each step."""
top_k: int = 1
"""The number of highest probability vocabulary tokens to
keep for top-k-filtering."""
repetition_penalty: float = 1.0
"""Penalizes repeated tokens according to frequency."""
random_seed: int = 0
"""The model generates random results.
Changing the random seed alone will produce a different response
with similar characteristics. It is possible to reproduce results
by fixing the random seed (assuming all other hyperparameters
are also fixed)"""
beam_search_diversity_rate: float = 1.0
"""Only applies to beam search, i.e. when the beam width is >1.
A higher value encourages beam search to return a more diverse
set of candidates"""
beam_width: Optional[int] = None
"""The number of concurrent candidates to keep track of during
beam search"""
length_pentaly: float = 1.0
"""Only applies to beam search, i.e. when the beam width is >1.
Larger values penalize long candidates more heavily, thus preferring
shorter candidates"""
writer_api_key: Optional[str] = None
stop: Optional[List[str]] = None
"""Sequences when completion generation will stop"""
base_url: Optional[str] = None
"""Base url to use, if None decides based on model name."""
class Config:
"""Configuration for this pydantic object."""
extra = Extra.forbid
@root_validator()
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key exists in environment."""
writer_api_key = get_from_dict_or_env(
values, "writer_api_key", "WRITER_API_KEY"
)
values["writer_api_key"] = writer_api_key
return values
@property
def _default_params(self) -> Mapping[str, Any]:
"""Get the default parameters for calling Writer API."""
return {
"tokens_to_generate": self.tokens_to_generate,
"stop": self.stop,
"logprobs": self.logprobs,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
"repetition_penalty": self.repetition_penalty,
"random_seed": self.random_seed,
"beam_search_diversity_rate": self.beam_search_diversity_rate,
"beam_width": self.beam_width,
"length_pentaly": self.length_pentaly,
}
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{"model_id": self.model_id}, **self._default_params}
@property
def _llm_type(self) -> str:
"""Return type of llm."""
return "writer"
def _call(self, prompt: str, stop: Optional[List[str]] = None) -> str:
"""Call out to Writer's complete endpoint.
Args:
prompt: The prompt to pass into the model.
stop: Optional list of stop words to use when generating.
Returns:
The string generated by the model.
Example:
.. code-block:: python
response = Writer("Tell me a joke.")
"""
if self.base_url is not None:
base_url = self.base_url
else:
base_url = (
"https://api.llm.writer.com/v1/models/{self.model_id}/completions"
)
response = requests.post(
url=base_url,
headers={
"Authorization": f"Bearer {self.writer_api_key}",
"Content-Type": "application/json",
"Accept": "application/json",
},
json={"prompt": prompt, **self._default_params},
)
text = response.text
if stop is not None:
# I believe this is required since the stop tokens
# are not enforced by the model parameters
text = enforce_stop_tokens(text, stop)
return text