config.py

from truss_train import definitions
from truss.base import truss_config


"""
Runtime provides runtime options for the training job. See the docs to learn more 
about configuring Training Cache and Automatic Checkpointing.
"""
runtime = definitions.Runtime(
    start_commands=[
        "/bin/sh -c './run.sh'",
    ],
    environment_variables={
        # Make sure these secrets are set in your Baseten Workspace
        "HF_TOKEN": definitions.SecretReference(name="hf_access_token"),
        "WANDB_API_KEY": definitions.SecretReference(name="wandb_api_key"),
        "BASE_MODEL_ID": "google/gemma-3-27b-it",
        "OUTPUT_LORA_REPO_ID": "rayraycano/finetune-demo-lora", # TODO: your HF Repo ID
    },
    enable_cache=True,
    # checkpointing_config=definitions.CheckpointingConfig(
    #     enabled=True,
    # ),
)

"""
Compute allows you to specify the hardware required for the training job. See the docs to learn more
about configuring multinode training.
"""
compute = definitions.Compute(
    accelerator=truss_config.AcceleratorSpec(
        accelerator=truss_config.Accelerator.H200,
        count=8,
    ),
    node_count=2,
)

"""
TrainingJob is the main configuration object for your training job. It includes the compute, runtime, and image.
"""
training_job = definitions.TrainingJob(
    compute=compute,
    runtime=runtime,
    # axolotl image includes most of the dependencies you need for training
    image=definitions.Image(base_image="axolotlai/axolotl:main-20250324-py3.11-cu124-2.6.0"),
)

"""
TrainingProject is an organizational tool to group your training jobs.
"""
first_project = definitions.TrainingProject(name="finetune-demo-full-feature-ori-dfw-2", job=training_job)