Skip to content

python-ml

Recipe card from the charly-languages plugin (Images — the deployable catalog).

GPU-accelerated Python environment with ML libraries — PyTorch, transformers, vLLM, llama.cpp.

Property Value
Base nvidia
Candies agent-forwarding, python-ml
Platforms linux/amd64
Registry ghcr.io/opencharly

The python-ml candy is a Tier 2 environment-owner meta-layer that:

  1. Owns the pixi.toml (core ML Python environment)
  2. Composes llama-cpp (Tier 1 sub-candy) via candy: [llama-cpp]
  3. Installs vLLM 0.19 wheel via a cmd task (after pixi env is established)

Build order: pixi environment → llama-cpp (binaries) → vLLM 0.19 wheel

  1. fedoranvidia (CUDA base)
  2. pixipython (transitive)
  3. python-ml — ML pixi environment (Tier 2, owns pixi.toml)
  4. llama-cpp — llama.cpp binaries (Tier 1, via candy: field)
Terminal window
charly box build python-ml
charly shell python-ml
# python -c "import torch; print(torch.cuda.is_available())"

After charly box build:

  • charly shell python-ml -c "python -c 'import torch; print(torch.cuda.is_available())'" — CUDA OK
  • charly shell python-ml -c "python -c 'import vllm; print(vllm.__version__)'" — vLLM OK
  • charly shell python-ml -c "ls ~/llama.cpp/llama-quantize" — llama.cpp OK

MUST be invoked when the task involves the python-ml box, ML training environments, or GPU-accelerated Python. Invoke this skill BEFORE reading source code or launching Explore agents.

  • /charly-image:image — image family umbrella (candy: image entries — those carrying base:/from: — in charly.yml, build/validate/inspect/list)
  • /charly-build:build — the embedded build vocabulary (distros, builders, init-systems)