id: bitnet version: 1.0.0 title: BitNet LLM description: short: Run BitNet LLM inference on CPU long: | BitNet is a CPU-based large language model inference engine from Microsoft. This package runs the bitnet-b1.58-2B-4T model and provides SSH access for interactive use. Requires AVX2 CPU support (Intel Haswell or newer). release-notes: | Initial release with SSH access enabled. license: Apache-2.0 wrapper-repo: https://github.com/kth8/bitnet upstream-repo: https://github.com/microsoft/BitNet support-site: https://github.com/kth8/bitnet/issues # No dependencies on other packages dependencies: {} # Docker image configuration containers: main: image: main mounts: main: /root # Volume for persistent data (models, etc.) volumes: main: type: data # Network interfaces interfaces: ssh: name: SSH description: SSH access to BitNet container tor-config: port-mapping: 22: "22" lan-config: 443: ssl: false internal: 22 ui: false protocols: - tcp - http # Configuration options config: get: type: script set: type: script properties: type: script # Health check - verify sshd is running health-checks: main: name: SSH Service success-message: SSH server is running and ready for connections type: script # Actions for the user actions: run-inference: name: Run Inference description: Run a test inference query warning: null allowed-statuses: - running implementation: type: docker image: main system: false entrypoint: python3 args: - /BitNet/run_inference.py - -m - ggml-model-i2_s.gguf - -p - "What is the meaning of life?" mounts: main: /root io-format: json backup: create: type: docker image: compat system: true entrypoint: compat args: - duplicity - create - /mnt/backup - /root/data mounts: BACKUP: /mnt/backup main: /root/data restore: type: docker image: compat system: true entrypoint: compat args: - duplicity - restore - /mnt/backup - /root/data mounts: BACKUP: /mnt/backup main: /root/data migrations: from: {} to: {}