Compatibility Matrix

Runtime requirements

Component	Minimum	Tested
Linux	Ubuntu 22.04	Ubuntu 24.04
WSL2	WSL2 on Windows 10/11	Windows 10 22H2
Rust toolchain	1.85.0	1.85.0 (pinned)
Python	3.11	3.11
vLLM	0.9.0	0.21.0 (resolved in `uv.lock`)
NVIDIA driver	555.x	555.x
CUDA toolkit	12.6	12.6
CUDA Compute Capability	8.0 (A100)	8.0+

Build requirements

Tool	Version
`cargo` + `rustup`	Rust 1.85.0
`maturin`	latest (install via `pip install maturin`)
`uv`	0.4+
`nvcc`	12.6 (only for `--features cuda`)
`mdbook`	latest (only for docs)

Model compatibility

Models that have been verified to work with Meridian's phase detection:

Model family	Boundary detection	Config
DeepSeek-R1	Token IDs `[128799, 128800]`	`models/deepseek_r1.toml`
Qwen3 / Qwen2.5	Token IDs `[151648, 151649]`	`models/qwen3.toml`
IBM Granite 3.2	Prose markers (no distinct token IDs)	`models/granite_3_2.toml`

Models that are not verified to work:

Models with non-standard <think> tokenisation not listed above — configure think_start_token_ids / think_end_token_ids manually and validate with a sample prompt before production use.
Models served through streaming APIs (e.g. Claude via Anthropic API) — Meridian requires direct access to the logit vector, which API-served models do not expose.

Feature flags

Feature flag	Requires	Status
(default — no flags)	Linux, Rust	Fully CI-tested
`--features prometheus`	`prometheus` crate	CI-tested
`--features unstable`	Rust nightly-gated APIs	CI-tested
`--features nixl`	`libnixl.so` on deploy host	Compiles; integration tested with synthetic mock
`--features cuda`	`nvcc`, CUDA 12.6, GPU at runtime	Build-tested on GPU CI runner

CI coverage

Job	Platform	GPU	Status
`rust-core`	ubuntu-24.04	No
`rust-kernels` (stub)	ubuntu-24.04	No	Same badge
`python`	ubuntu-24.04	No	Same badge
`docs`	ubuntu-24.04	No	Same badge
`supply-chain`	ubuntu-24.04	No	Same badge
`cuda-build`	self-hosted `gpu` runner	Yes	Runs only on `angelnicolasc` org pushes

The GPU jobs are gated to prevent arbitrary code execution on the self-hosted runner from fork PRs. See GPU CI runner setup.

Known incompatibilities

vLLM below 0.9.0: the dependency constraint is vllm>=0.9.0. Earlier versions are not supported and will be rejected at install time.
Windows (native): the Rust workspace builds on Windows (tested in development), but the Python extension and benchmarks require Linux or WSL2 for the CUDA and maturin paths.
macOS: not supported. CUDA is not available on macOS.