mycellm
Copyright (c) mycellm contributors

This product includes software developed as part of the mycellm project,
licensed under the Apache License, Version 2.0.

---

Third-party attributions:

* The continuous-batching MLX backend (src/mycellm/inference/mlx_batched.py)
  and the batching benchmark (scripts/bench_mlx_batching.py) were informed by
  the engine design of oMLX (https://github.com/jundot/omlx), in particular
  omlx/engine/batched.py, which is licensed under the Apache License,
  Version 2.0. mycellm's implementation drives mlx-lm's BatchGenerator directly
  and does not copy oMLX source; oMLX is credited as a design reference.

* Several 0.6.3 features were informed by the engineering write-ups in the
  oMLX v0.5.4rc1 release (https://github.com/jundot/omlx, Apache-2.0): the
  empirical context-memory calibration benchmark
  (scripts/bench_context_calibration.py), the runtime memory-pressure
  watcher (src/mycellm/memory_watch.py), the streaming stop-sequence
  holdback and truncated tool-call recovery, exact streaming usage
  reporting, and length-grouped embedding batching
  (src/mycellm/inference/mlx_embed.py). mycellm's implementations are
  original and do not copy oMLX source; oMLX is credited as a design
  reference.

* The verified model-download pipeline (src/mycellm/inference/hf_verify.py
  and its iOS mirror) was informed by the manifest/hash-verified streaming
  installer described in turbo-fieldfare
  (https://github.com/drumih/turbo-fieldfare). No source was copied;
  turbo-fieldfare is credited as a design reference.
