requirements.yaml reference

Beta

The AI Runtime CLI is in Beta.

requirements.yaml defines the Python libraries installed for your workload. It is referenced from the workload YAML's environment block:

YAML

environment:
  dependencies: requirements.yaml

The file format follows the Databricks Base Environment Specification.

Example

YAML

environment_version: '4'
dependencies:
  - --index-url https://pypi.org/simple
  - -r '/Workspace/Shared/requirements.txt'
  - my-library==6.1
  - /Workspace/Shared/Path/To/simplejson-3.19.3-py3-none-any.whl

In the example above:

1. The Base Environment v4 image is used.
2. A requirements.txt referenced in the Workspace is installed.
3. my-library is installed at version 6.1.
4. A wheel in the Workspace is installed.

Reuse an existing requirements.txt

If you already maintain a requirements.txt, reference it directly from your Workspace:

YAML

version: '4'
dependencies:
  - -r '/Workspace/path/to/requirements.txt'

The version field is optional and defaults to "4" if not specified.

Axolotl example

Dependency file for training with Axolotl:

YAML

version: '4'
dependencies:
  # Standard build tools and MLflow
  - packaging
  - setuptools
  - wheel
  - ninja
  - mlflow>=3.6

  # Axolotl with flash-attn and deepspeed (no build isolation)
  - --no-build-isolation
  - axolotl[flash-attn,deepspeed]

  # Flash attention wheel (no deps — specific CUDA / torch version)
  - --no-deps
  - https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp312-cp312-linux_x86_64.whl