Welcome to Demo for InstructLab CLI¶

A tool to help you contribute to InstructLab Taxonomy.

It allows you to locally

Check the behavior of the latest Merlinite-7B model
Iterate seed examples for synthetic data generation
Validate the synthetic data via local LoRA training

This demo walks through the functionality step by step.

For more info, see the CLI repo.

Step 0: Initial setup¶

Create workspace and virtual environment

python3 -m venv workspace/venv
cd workspace
source venv/bin/activate.fish

Install CLI

pip install instructlab

Initialize workspace

ilab config init

Step 1: Check current model behavior¶

Download current model

ilab model download

Start chat

ilab model chat

Trying…

how to initialize a workspace using the CLI of InstructLab?

…but the answer is wrong

Step 2: Write seed examples for SDG¶

Check current status of taxonomy

ilab taxonomy diff

Check YAML file to add

cat ../qna.yaml

Add file to taxonomy and check updated status of taxonomy

mkdir -p taxonomy/knowledge/instructlab/cli
cp ../qna.yaml taxonomy/knowledge/instructlab/cli/qna.yaml
ilab taxonomy diff

Step 3: Generate synthetic data¶

Generate 25 synthetic samples

ilab data generate --num-instructions 25

Here we use 25 for demo purpose only.
Usually we recommend more samples (e.g. 100).

It takes a while, feel free to fast-forward if you are watching the recording.

Check output

ls
ls generated

Step 4: Update model via LoRA training¶

Perform Q-LoRA for 100 iterations

ilab model train --iters 100

Here we use 100 for this particular demo.
The actual iterations needed depends on the complexity of the data.

It takes a while, feel free to fast-forward if you are watching the recording.

Check output

ls
ls instructlab-merlinite-7b-lab-mlx-q

Step 5: Check before/after behavior¶

Run old/new model on test data

ilab model test

Step 6: Interact with updated model¶

Convert updated model to GGUF

ilab model convert --model-dir instructlab-merlinite-7b-lab-mlx-q

Serve updated model and chat

ilab model serve --model-path instructlab-merlinite-7b-lab-mlx-q-fused-pt/*-Q4_K_M.gguf &
ilab model chat

Trying again…

how to initialize a workspace using the CLI of InstructLab?

…and the answer is now correct!

Step -1: Make contribution to taxonomy¶

Commit change and push to fork

cd taxonomy
git status
git remote add xukai92 https://github.com/instructlab/taxonomy.git
git add knowledge/instructlab/cli
git commit -sm "feat(knowledge): InstructLab CLI usage"
git checkout -b demo
git push --set-upstream xukai92 demo

Now, you can go creating a pull request on the taxonomy repository!

–> instructlab/taxonomy

For more info, see the CLI repo.