Beam
调用 Beam API 包装器以在云部署中部署并对 gpt2 LLM 实例进行后续调用。需要安装 Beam 库并注册 Beam 客户端 ID 和客户端密钥。通过调用包装器创建并运行模型的实例，返回的文本与提示相关。然后可以通过直接调用 Beam API 进行额外的调用。
创建一个账户，如果你还没有的话。从 仪表板 获取你的 API 密钥。
安装 Beam CLI
!curl https://raw.githubusercontent.com/slai-labs/get-beam/main/get-beam.sh -sSfL | sh
注册API密钥并设置您的Beam客户端ID和秘密环境变量：
import os
beam_client_id = "<Your beam client id>"
beam_client_secret = "<Your beam client secret>"
# Set the environment variables
os.environ["BEAM_CLIENT_ID"] = beam_client_id
os.environ["BEAM_CLIENT_SECRET"] = beam_client_secret
# Run the beam configure command
!beam configure --clientId={beam_client_id} --clientSecret={beam_client_secret}
安装Beam SDK：
%pip install --upgrade --quiet beam-sdk
直接从LangChain部署并调用Beam！
请注意，冷启动可能需要几分钟才能返回响应，但后续调用会更快！
<!--IMPORTS:[{"imported": "Beam", "source": "langchain_community.llms.beam", "docs": "https://python.langchain.com/api_reference/community/llms/langchain_community.llms.beam.Beam.html", "title": "Beam"}]-->
from langchain_community.llms.beam import Beam
llm = Beam(
model_name="gpt2",
name="langchain-gpt2-test",
cpu=8,
memory="32Gi",
gpu="A10G",
python_version="python3.8",
python_packages=[
"diffusers[torch]>=0.10",
"transformers",
"torch",
"pillow",
"accelerate",
"safetensors",
"xformers",
],
max_length="50",
verbose=False,
)
llm._deploy()
response = llm._call("Running machine learning on a remote GPU")
print(response)
