Predibase

Predibase 讓您訓練、微調和部署任何 ML 模型，從線性迴歸到大型語言模型。

此範例示範如何將 Langchain 與部署在 Predibase 上的模型一起使用

設定 (Setup)

若要執行此筆記本，您需要一個 Predibase 帳戶 和一個 API 金鑰。

您還需要安裝 Predibase Python 套件

%pip install --upgrade --quiet  predibase
import os

os.environ["PREDIBASE_API_TOKEN"] = "{PREDIBASE_API_TOKEN}"

初始呼叫 (Initial Call)

from langchain_community.llms import Predibase

model = Predibase(
    model="mistral-7b",
    predibase_api_key=os.environ.get("PREDIBASE_API_TOKEN"),
)

API 參考：Predibase

from langchain_community.llms import Predibase

# With a fine-tuned adapter hosted at Predibase (adapter_version must be specified).
model = Predibase(
    model="mistral-7b",
    predibase_api_key=os.environ.get("PREDIBASE_API_TOKEN"),
    predibase_sdk_version=None,  # optional parameter (defaults to the latest Predibase SDK version if omitted)
    adapter_id="e2e_nlg",
    adapter_version=1,
    **{
        "api_token": os.environ.get("HUGGING_FACE_HUB_TOKEN"),
        "max_new_tokens": 5,  # default is 256
    },
)

API 參考：Predibase

from langchain_community.llms import Predibase

# With a fine-tuned adapter hosted at HuggingFace (adapter_version does not apply and will be ignored).
model = Predibase(
    model="mistral-7b",
    predibase_api_key=os.environ.get("PREDIBASE_API_TOKEN"),
    predibase_sdk_version=None,  # optional parameter (defaults to the latest Predibase SDK version if omitted)
    adapter_id="predibase/e2e_nlg",
    **{
        "api_token": os.environ.get("HUGGING_FACE_HUB_TOKEN"),
        "max_new_tokens": 5,  # default is 256
    },
)

API 參考：Predibase

# Optionally use `kwargs` to dynamically overwrite "generate()" settings.
response = model.invoke(
    "Can you recommend me a nice dry wine?",
    **{"temperature": 0.5, "max_new_tokens": 1024},
)
print(response)

鏈呼叫設定 (Chain Call Setup)

from langchain_community.llms import Predibase

model = Predibase(
    model="mistral-7b",
    predibase_api_key=os.environ.get("PREDIBASE_API_TOKEN"),
    predibase_sdk_version=None,  # optional parameter (defaults to the latest Predibase SDK version if omitted)
    **{
        "api_token": os.environ.get("HUGGING_FACE_HUB_TOKEN"),
        "max_new_tokens": 5,  # default is 256
    },
)

API 參考：Predibase

# With a fine-tuned adapter hosted at Predibase (adapter_version must be specified).
model = Predibase(
    model="mistral-7b",
    predibase_api_key=os.environ.get("PREDIBASE_API_TOKEN"),
    predibase_sdk_version=None,  # optional parameter (defaults to the latest Predibase SDK version if omitted)
    adapter_id="e2e_nlg",
    adapter_version=1,
    **{
        "api_token": os.environ.get("HUGGING_FACE_HUB_TOKEN"),
        "max_new_tokens": 5,  # default is 256
    },
)

# With a fine-tuned adapter hosted at HuggingFace (adapter_version does not apply and will be ignored).
llm = Predibase(
    model="mistral-7b",
    predibase_api_key=os.environ.get("PREDIBASE_API_TOKEN"),
    predibase_sdk_version=None,  # optional parameter (defaults to the latest Predibase SDK version if omitted)
    adapter_id="predibase/e2e_nlg",
    **{
        "api_token": os.environ.get("HUGGING_FACE_HUB_TOKEN"),
        "max_new_tokens": 5,  # default is 256
    },
)

SequentialChain

from langchain.chains import LLMChain
from langchain_core.prompts import PromptTemplate

API 參考：LLMChain | PromptTemplate

# This is an LLMChain to write a synopsis given a title of a play.
template = """You are a playwright. Given the title of play, it is your job to write a synopsis for that title.

Title: {title}
Playwright: This is a synopsis for the above play:"""
prompt_template = PromptTemplate(input_variables=["title"], template=template)
synopsis_chain = LLMChain(llm=llm, prompt=prompt_template)

# This is an LLMChain to write a review of a play given a synopsis.
template = """You are a play critic from the New York Times. Given the synopsis of play, it is your job to write a review for that play.

Play Synopsis:
{synopsis}
Review from a New York Times play critic of the above play:"""
prompt_template = PromptTemplate(input_variables=["synopsis"], template=template)
review_chain = LLMChain(llm=llm, prompt=prompt_template)

# This is the overall chain where we run these two chains in sequence.
from langchain.chains import SimpleSequentialChain

overall_chain = SimpleSequentialChain(
    chains=[synopsis_chain, review_chain], verbose=True
)

API 參考：SimpleSequentialChain

review = overall_chain.run("Tragedy at sunset on the beach")

微調的 LLM（使用您自己來自 Predibase 的微調 LLM）(Fine-tuned LLM (Use your own fine-tuned LLM from Predibase))

from langchain_community.llms import Predibase

model = Predibase(
    model="my-base-LLM",
    predibase_api_key=os.environ.get(
        "PREDIBASE_API_TOKEN"
    ),  # Adapter argument is optional.
    predibase_sdk_version=None,  # optional parameter (defaults to the latest Predibase SDK version if omitted)
    adapter_id="my-finetuned-adapter-id",  # Supports both, Predibase-hosted and HuggingFace-hosted adapter repositories.
    adapter_version=1,  # required for Predibase-hosted adapters (ignored for HuggingFace-hosted adapters)
    **{
        "api_token": os.environ.get("HUGGING_FACE_HUB_TOKEN"),
        "max_new_tokens": 5,  # default is 256
    },
)
# replace my-base-LLM with the name of your choice of a serverless base model in Predibase

API 參考：Predibase

# Optionally use `kwargs` to dynamically overwrite "generate()" settings.
# response = model.invoke("Can you help categorize the following emails into positive, negative, and neutral?", **{"temperature": 0.5, "max_new_tokens": 1024})

相關 (Related)

• LLM 概念指南
• LLM 操作指南