Scribd
Upload
71 views

Instructlab Cheatsheet

Cheatsheet for using Instructlab

Uploaded by

Nicholas Renotte
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
71 views

Instructlab Cheatsheet

Cheatsheet for using Instructlab

Uploaded by

Nicholas Renotte
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

INSTRUCTLAB CHEATSHEET

// 1. Install dependencies and clone the // 6. Generate synthetic data pairs using LLM
taxonomy # This will use the LLM specified in the config
# Install the instructlab library file to generate synthetic instruction pairs, by
pip install instructlab default it’ll create 100 examples but this can be
# The taxonomy gives you a structured format changed. MAKE SURE TO VALIDATE THE DATA

TE
for managing your training data BEFORE TRAINING...PLEASE ! 🥲
git clone ilab data generate
https://github.com/instructlab/taxonomy

OT
// 7. train the model
// 2. Initialize Instructlab # This will kick off the training of a new LoRA for
# This creates a baseline config.yaml file where and will save the trained adapter in the
you can specify which models to use for instructlab-merlinite-7b-lab-trained directory.
training, synethetic data generation and serving ilab model train

N
ilab config init # If you want to change how long you train the
model for you can set the --iters flag e.g.

RE
// 3. Download baseline modelmodels training for 1000 iterations.
# Note that the base model that this will ilab model train --iters 1000
download is defined in the config # Also if you get stuck and need help you can
ilab model download run the cli command with the --help flag
ilab model train --help
S
// 4. Spin up the API and chat with it
# Run this in one terminal to host the API using // 8. Quantize the Model
LA

an OpenAI compatible web server # This automatically quantizes your model for
ilab model serve more efficient serving and inference.
# Chat with it using another terminal, note you ilab model convert
can also hit the API directly
HO

ilab model chat // 9. Serve and chat with the trained model
# Serve the trained model by specifying the
// 5. Upload your dataset into a markdown math to the quantized model
format and validate ilab model serve --model-path instructlab-
# I used GitHub, this is an example: merlinite-7b-lab-trained/instructlab-merlinite-
IC

https://github.com/nicknochnack/testdata This 7b-lab-Q4_K_M.gguf


site worked pretty well to do the conversion
https://pdf2md.morethan.io/. Then setup a yaml # Chat with the new model
N

file that points to the dataset example YAML: ilab model chat -gm -m instructlab-merlinite-7b-
https://github.com/nicknochnack/testdata/blob lab-trained/instructlab-merlinite-7b-lab-
/main/qna.yaml Q4_K_M.gguf

# You can view and validate the changes to your


taxonomy by running the diff command
ilab taxonomy diff

NICHOLAS RENOTTE

You might also like