Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

mlx-beam

B.E.A.M. — Batched Engine for Apple Metal. An MLX inference engine for hybrid-attention and mixture-of-experts language models on Apple silicon.

Work in progress. See the status table below and the changelog.

What it is for

Models like Qwen3.8 (Gated DeltaNet + attention) and Qwen3.8-Flash-Next (125B-A6B MoE with an on-disk n-gram table) do not fit the assumptions of a plain transformer server: their recurrent state cannot be trimmed like a KV cache, their experts do not fit in memory, and speculative decoding under batching needs a rollback that most engines lack. B.E.A.M. is built around those cases:

  • a prefix cache with recurrent-state checkpoints, so a long conversation resumes from disk instead of re-prefilling;
  • multi-token prediction inside continuous batching;
  • expert streaming from SSD with a resident pool, for models larger than RAM;
  • a small, explicit OpenAI-compatible API, with everything else as optional extras.

The engine reads standard MLX checkpoints and the B.E.A.M. package layout (extras/ next to the shards; see the model cards under huggingface.co/p4ik).

Install

uv tool install "mlx-beam @ git+https://github.com/p4ik/mlx-beam"
beam doctor

Inside an existing uv project: uv add "mlx-beam @ git+https://github.com/p4ik/mlx-beam", then uv run beam doctor.

beam doctor prints the Python, MLX, device and memory it sees (--json for scripts) and exits non-zero when MLX is missing or fails to load. It is the only command so far.

Status

Piece State
CLI, packaging, CI skeleton
vendored mlx-lm base planned
prefix cache with recurrent-state checkpoints planned
MTP in the batch planned
expert streaming planned

Contributing

See CONTRIBUTING.md. Working rules for people and agents are in AGENTS.md.

License

Apache-2.0. Vendored components keep their own licenses; see NOTICE and VENDORED.md.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

mlx_beam-0.1.0.dev5.tar.gz (8.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

mlx_beam-0.1.0.dev5-py3-none-any.whl (9.1 kB view details)

Uploaded Python 3

File details

Details for the file mlx_beam-0.1.0.dev5.tar.gz.

File metadata

  • Download URL: mlx_beam-0.1.0.dev5.tar.gz
  • Upload date:
  • Size: 8.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for mlx_beam-0.1.0.dev5.tar.gz
Algorithm Hash digest
SHA256 8038bf7c4c3b4494d7ab26df8342eaa9e3c183f2564766e42d7b0f802d51563b
MD5 bf01d0b9480c428fdda3f71d7d8741e8
BLAKE2b-256 dc2042ca91e031de8ac019d837d20c3262bd552a7f62a294f760168062f734c5

See more details on using hashes here.

Provenance

The following attestation bundles were made for mlx_beam-0.1.0.dev5.tar.gz:

Publisher: release.yml on p4ik/mlx-beam

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file mlx_beam-0.1.0.dev5-py3-none-any.whl.

File metadata

  • Download URL: mlx_beam-0.1.0.dev5-py3-none-any.whl
  • Upload date:
  • Size: 9.1 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for mlx_beam-0.1.0.dev5-py3-none-any.whl
Algorithm Hash digest
SHA256 be8d6d708c44a1a8b2330a893e0303899f8a25b971b6b4cdc9851326b591c502
MD5 83d10d1bdf03eb32cb76c128c00f2d39
BLAKE2b-256 3408d7abe0159164dfcb3ea4b9fd3e3f8d9a11ea047419f2f632991ecef712b3

See more details on using hashes here.

Provenance

The following attestation bundles were made for mlx_beam-0.1.0.dev5-py3-none-any.whl:

Publisher: release.yml on p4ik/mlx-beam

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Release history Release notifications | RSS feed

This release

0.1.0.dev5 This release

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page