SimTsai / langchain-ChatGLM

langchain-ChatGLM

esay use ChatGLM in LangChain.

ChatGLM2-6B langchain

Install Requirement

pip install -r requirements.txt

Usage

from chatglm_pipline import ChatGLMPipeline
from langchain import PromptTemplate, LLMChain
from langchain.callbacks.manager import CallbackManager
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler

callback_manager = CallbackManager([StreamingStdOutCallbackHandler()])

llm = ChatGLMPipeline.from_model_id(
    model_id="THUDM/chatglm2-6b",
    device=-1, # if use GPU set to 0
    model_kwargs={"temperature": 0, "max_length": 64, "trust_remote_code": True},
    callback_manager=callback_manager, 
    verbose=True,
)

template = """问: {question}

答: 让我们一步一步地思考."""

prompt = PromptTemplate(template=template, input_variables=["question"])

llm_chain = LLMChain(prompt=prompt, llm=llm)

question = "华晨宇和张碧晨是什么关系?"

llm_chain.run(question)

model_kwargs

key	values	remark
"device"	"cuda"	使用cuda加速
"float"	True, False	使用cpu推理
"quantize"	8	量化