forked from Archives/langchain
You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
120 lines
3.8 KiB
Python
120 lines
3.8 KiB
Python
"""Wrapper around Cohere APIs."""
|
|
import os
|
|
from typing import Any, Dict, List, Optional
|
|
|
|
from pydantic import BaseModel, Extra, root_validator
|
|
|
|
from langchain.llms.base import LLM, CompletionOutput
|
|
from langchain.llms.utils import enforce_stop_tokens
|
|
|
|
|
|
class Cohere(BaseModel, LLM):
|
|
"""Wrapper around Cohere large language models.
|
|
|
|
To use, you should have the ``cohere`` python package installed, and the
|
|
environment variable ``COHERE_API_KEY`` set with your API key.
|
|
|
|
Example:
|
|
.. code-block:: python
|
|
|
|
from langchain import Cohere
|
|
cohere = Cohere(model="small")
|
|
"""
|
|
|
|
client: Any #: :meta private:
|
|
model: str = "small"
|
|
"""Model name to use."""
|
|
|
|
max_tokens: int = 256
|
|
"""Denotes the number of tokens to predict per generation."""
|
|
|
|
temperature: float = 0.75
|
|
"""A non-negative float that tunes the degree of randomness in generation."""
|
|
|
|
k: int = 0
|
|
"""Number of most likely tokens to consider at each step."""
|
|
|
|
p: int = 1
|
|
"""Total probability mass of tokens to consider at each step."""
|
|
|
|
frequency_penalty: int = 0
|
|
"""Penalizes repeated tokens according to frequency."""
|
|
|
|
presence_penalty: int = 0
|
|
"""Penalizes repeated tokens."""
|
|
|
|
num_generations: int = 1
|
|
"""Number of generations to return."""
|
|
|
|
return_likelihoods: bool = True
|
|
"""Whether to return the likelihoods of the generated tokens."""
|
|
|
|
class Config:
|
|
"""Configuration for this pydantic object."""
|
|
|
|
extra = Extra.forbid
|
|
|
|
@root_validator()
|
|
def validate_environment(cls, values: Dict) -> Dict:
|
|
"""Validate that api key and python package exists in environment."""
|
|
if "COHERE_API_KEY" not in os.environ:
|
|
raise ValueError(
|
|
"Did not find Cohere API key, please add an environment variable"
|
|
" `COHERE_API_KEY` which contains it."
|
|
)
|
|
try:
|
|
import cohere
|
|
|
|
values["client"] = cohere.Client(os.environ["COHERE_API_KEY"])
|
|
except ImportError:
|
|
raise ValueError(
|
|
"Could not import cohere python package. "
|
|
"Please it install it with `pip install cohere`."
|
|
)
|
|
return values
|
|
|
|
def generate(self, prompt: str, stop: Optional[List[str]] = None) -> List[CompletionOutput]:
|
|
"""Call out to Cohere's generate 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 = cohere("Tell me a joke.")
|
|
"""
|
|
response = self.client.generate(
|
|
model=self.model,
|
|
prompt=prompt,
|
|
max_tokens=self.max_tokens,
|
|
temperature=self.temperature,
|
|
k=self.k,
|
|
p=self.p,
|
|
frequency_penalty=self.frequency_penalty,
|
|
presence_penalty=self.presence_penalty,
|
|
stop_sequences=stop,
|
|
num_generations=self.num_generations,
|
|
return_likelihoods="GENERATION" if self.return_likelihoods else None,
|
|
)
|
|
results = []
|
|
for generation in response.generations:
|
|
txt = generation.text
|
|
if stop is not None:
|
|
# If stop tokens are provided, Cohere's endpoint returns them.
|
|
# In order to make this consistent with other endpoints, we strip them.
|
|
txt = enforce_stop_tokens(txt, stop)
|
|
N = len(generation.token_likelihoods)
|
|
logprobs = [token.likelihood / N for token in generation.token_likelihoods]
|
|
results.append(
|
|
CompletionOutput(
|
|
text=txt,
|
|
logprobs=logprobs,
|
|
)
|
|
)
|
|
return results
|