Skip to main content
Version: 0.4.3

Model YAML

The model.yaml file is the heart of every Constellaxion project. It defines what model you’re using, what dataset to train on (if any), and where to deploy it.

After you run:

constellaxion init

Once initialized, Constellaxion generates a job.json file based on your model.yaml.

📂 Example: Deployment Only (No Training)​

model:
  id: cxn-foundation-model
  base: tiiuae/falcon-7b-instruct
deploy:
  aws:
    region: us-east-1

📂 Example: Fine-Tuning​

model:
  id: crypto-sentiment-v2
  base: tiiuae/falcon-7b-instruct
dataset:
  train: ./train.csv
  val: ./val.csv
  test: ./test.csv
training:
  epochs: 1
  batch_size: 16
deploy:
  aws:
    region: eu-west-2

This version includes local dataset files and training configuration for fine-tuning a base model.

🔍 Field Reference​

model​

id string - The unique label for this project or experiment.

base string - The Hugging Face model ID used as the starting point for deployment or fine-tuning.

dataset (required if fine-tuning)​

  • Paths to your local training, validation, and test datasets (CSV format)
  • All CSVs must have two columns: prompt and response
dataset:
  train: ./train.csv
  val: ./val.csv
  test: ./test.csv

training (required if fine-tuning)​

Configuration for fine-tuning. This section is required if dataset is present.

training:
  epochs: 1
  batch_size: 16

You can extend this section with additional parameters in the future (e.g., learning rate, optimizer, etc.)

deploy​

Describes where and how to deploy the model.

deploy:
  gcp:
    project_id: my-gcp-project
    region: europe-west2

project_id string (required if gcp) - Your GCP project ID

region string - Region where the model will be deployed

📌 Notes​

  • model.yaml is user-authored and committed to version control
  • job.json is generated from it and should not be edited manually
  • If you edit model.yaml, run constellaxion init again to refresh your project state

🔗 Related Pages​