Arcee

:tulip: The open source alignment toolkit for finetuning and deploying LLMs :tulip:

Finetuning Datasets

The Arcee toolkit contains routines for you to manage and generate finetuning datasets. The arcee toolkit supports supervised finetuning (SFT) aka intruction tuning.

✍️ Instruction Set ✍️

Instruction tuning datasets contain a series of prompt and completion examples.

Instruction sets can be loaded from a csv.

from arcee.data import InstructionSet
instruction_set = InstructionSet("./datasets/strip_api.csv")

Or downloaded from the Arcee platform

from arcee.data import InstructionSet
instruction_set = InstructionSet("https://app.arcee.ai/arcee/alpaca")

Self-Instruct Generation

Evolv-Instruct Genertation

Explain-Instruct Generation

Finetuning Models

HuggingFace

from arcee.models import LM
from arcee.data import Instuctions

lm = LM("falcon30b")
instructions = Instructions("./datasets/stripe-api.json")
lm.train(instructions)

lm.predict("Place an order for the LLM-9000 product for 100 USD to the card 3007200039992000")

OpenAI

Cohere

Together

Mosaic ML

LangChain Integration

from langchain import Arcee
#goes in llms/arcee.py

prompt_template = "Write a stripe API request for the following: {order}."

llm = Arcee(temperature=0)
llm_chain = LLMChain(
    llm=llm,
    prompt=PromptTemplate.from_template(prompt_template)
)
llm_chain("Place an order for the LLM-9000 product for 100 USD to the card 3007200039992000")

Domain Pretraining

Coming soon!

🗻 Arcee Platform 🗻

Arcee offers a platform for managing your proprietary language models in production. We offer a version of our platform hosted in our cloud as well as a deployable version that you can take on prem.

Authenticaiton

To use the arcee platform, visit https://app.arcee.com/api-settings and export ARCEE_API_KEY=******* in your environment.

Dataset Managemnt

View, search, and edit datasets.

instruction_set.upload("project-name")

Hosted Training

Hosted Inference

Lifecycle Management