[AUR-389] Add base interface and embedding model (#17)

This change provides the base interface of an embedding, and wrap the Langchain's OpenAI embedding. Usage as follow:

```python
from kotaemon.embeddings import AzureOpenAIEmbeddings

model = AzureOpenAIEmbeddings(
    model="text-embedding-ada-002",
    deployment="embedding-deployment",
    openai_api_base="https://test.openai.azure.com/",
    openai_api_key="some-key",
)
output = model("Hello world")
```
This commit is contained in:
Nguyen Trung Duc (john)
2023-09-14 14:08:58 +07:00
committed by GitHub
parent 1061192731
commit c339912312
6 changed files with 4772 additions and 3 deletions

View File

@@ -0,0 +1,62 @@
from typing import List, Type
from langchain.embeddings.base import Embeddings as LCEmbeddings
from theflow import Param
from ..components import BaseComponent
from ..documents.base import Document
class Embeddings(BaseComponent):
...
class LangchainEmbeddings(Embeddings):
_lc_class: Type[LCEmbeddings]
def __init__(self, **params):
if self._lc_class is None:
raise AttributeError(
"Should set _lc_class attribute to the LLM class from Langchain "
"if using LLM from Langchain"
)
self._kwargs: dict = {}
for param in list(params.keys()):
if param in self._lc_class.__fields__: # type: ignore
self._kwargs[param] = params.pop(param)
super().__init__(**params)
def __setattr__(self, name, value):
if name in self._lc_class.__fields__:
setattr(self.agent, name, value)
else:
super().__setattr__(name, value)
@Param.decorate(no_cache=True)
def agent(self):
return self._lc_class(**self._kwargs)
def run_raw(self, text: str) -> List[float]:
return self.agent.embed_query(text) # type: ignore
def run_batch_raw(self, text: List[str]) -> List[List[float]]:
return self.agent.embed_documents(text) # type: ignore
def run_document(self, text: Document) -> List[float]:
return self.agent.embed_query(text.text) # type: ignore
def run_batch_document(self, text: List[Document]):
return self.agent.embed_documents([each.text for each in text]) # type: ignore
def is_document(self, text) -> bool:
if isinstance(text, Document):
return True
elif isinstance(text, List) and isinstance(text[0], Document):
return True
return False
def is_batch(self, text) -> bool:
if isinstance(text, list):
return True
return False

View File

@@ -0,0 +1,15 @@
from langchain.embeddings import OpenAIEmbeddings as LCOpenAIEmbeddings
from .base import LangchainEmbeddings
class OpenAIEmbeddings(LangchainEmbeddings):
_lc_class = LCOpenAIEmbeddings
class AzureOpenAIEmbeddings(LangchainEmbeddings):
_lc_class = LCOpenAIEmbeddings
def __init__(self, **params):
params["openai_api_type"] = "azure"
super().__init__(**params)

View File

@@ -30,8 +30,8 @@ class LangchainChatLLM(ChatLLM):
self._kwargs[param] = params.pop(param)
super().__init__(**params)
@Param.decorate()
def agent(self):
@Param.decorate(no_cache=True)
def agent(self) -> BaseLanguageModel:
return self._lc_class(**self._kwargs)
def run_raw(self, text: str) -> LLMInterface:
@@ -43,7 +43,7 @@ class LangchainChatLLM(ChatLLM):
return self.run_batch_document(inputs)
def run_document(self, text: List[Message]) -> LLMInterface:
pred = self.agent.generate([text])
pred = self.agent.generate([text]) # type: ignore
return LLMInterface(
text=[each.text for each in pred.generations[0]],
completion_tokens=pred.llm_output["token_usage"]["completion_tokens"],