Skip to main content

模态框 (Modal)

This page covers how to use the Modal ecosystem to run LangChain custom LLMs. It is broken into two parts:

  1. Modal安装和Web端点部署 (Modal installation and web endpoint deployment)
  2. 使用部署的Web端点和LLM包装类 (Using deployed web endpoint with LLM wrapper class)

安装和设置 (Installation and Setup)​

  • 使用 pip install modal 进行安装 (Install with pip install modal)
  • 运行 modal token new (Run modal token new)

定义你的Modal函数和Webhooks (Define your Modal Functions and Webhooks)​

你必须包含一个提示。有一个严格的响应结构 (You must include a prompt. There is a rigid response structure):

class Item(BaseModel):
    prompt: str

@stub.function()
@modal.web_endpoint(method="POST")
def get_text(item: Item):
    return {"prompt": run_gpt2.call(item.prompt)}

以下是一个使用GPT2模型的示例 (The following is an example with the GPT2 model):

from pydantic import BaseModel

import modal

CACHE_PATH = "/root/model_cache"

class Item(BaseModel):
    prompt: str

stub = modal.Stub(name="example-get-started-with-langchain")

def download_model():
    from transformers import GPT2Tokenizer, GPT2LMHeadModel
    tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
    model = GPT2LMHeadModel.from_pretrained('gpt2')
    tokenizer.save_pretrained(CACHE_PATH)
    model.save_pretrained(CACHE_PATH)

# 定义一个用于下面的LLM函数的容器镜像，该函数会下载并存储GPT-2模型 (Define a container image for the LLM function below, which downloads and stores the GPT-2 model)
image = modal.Image.debian_slim().pip_install(
    "tokenizers", "transformers", "torch", "accelerate"
).run_function(download_model)

@stub.function(
    gpu="any",
    image=image,
    retries=3,
)
def run_gpt2(text: str):
    from transformers import GPT2Tokenizer, GPT2LMHeadModel
    tokenizer = GPT2Tokenizer.from_pretrained(CACHE_PATH)
    model = GPT2LMHeadModel.from_pretrained(CACHE_PATH)
    encoded_input = tokenizer(text, return_tensors='pt').input_ids
    output = model.generate(encoded_input, max_length=50, do_sample=True)
    return tokenizer.decode(output[0], skip_special_tokens=True)

@stub.function()
@modal.web_endpoint(method="POST")
def get_text(item: Item):
    return {"prompt": run_gpt2.call(item.prompt)}

部署Web端点 (Deploy the web endpoint)​

使用 modal deploy CLI命令将Web端点部署到Modal云。 你的Web端点将在modal.run域名下获得一个持久的URL (Deploy the web endpoint to Modal cloud with the modal deploy CLI command. Your web endpoint will acquire a persistent URL under the modal.run domain).

LLM包装Modal Web端点 (LLM wrapper around Modal web endpoint)​

Modal LLM包装类将接受你部署的Web端点的URL (The Modal LLM wrapper class which will accept your deployed web endpoint's URL).

from langchain.llms import Modal

endpoint_url = "https://ecorp--custom-llm-endpoint.modal.run"  # 用你部署的Modal Web端点的URL替换我 (REPLACE ME with your deployed Modal web endpoint's URL)

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

question = "What NFL team won the Super Bowl in the year Justin Beiber was born?"

llm_chain.run(question)