jax-ml/jax

Choose this fork only if you need an old, frozen JAX baseline or existing fork-specific edits. For new work, current upstream is the better choice because this fork is far behind and likely missing many modern features and fixes.

Prefer upstream unless you specifically need this fork’s historical experimental features or older semantics. This fork looks like a stale, highly divergent snapshot with a few niche additions, not a maintained alternative platform.

Prefer this fork only if you need its TPU/GPU internals or experimental accelerator features. For general JAX use, upstream is a better default because this fork is materially behind and likely to lag on fixes, compatibility, and release-quality maintenance.

Prefer this fork only if you must keep an older JAX API surface alive. For new work, or anything expecting modern JAX features and active maintenance, upstream is the clear choice.

Choose this fork only if you need its exact experimental JAX behavior or historical accelerator patches. For most users, upstream JAX is the safer choice because this fork is far behind and appears to have lost or disabled substantial upstream functionality.

Prefer upstream unless you specifically need this fork's older experimental backend work or are continuing an existing dependency on its 2023-era codebase. For most adopters, the staleness and divergence are the main reasons to avoid it.

Prefer upstream JAX unless you specifically want this fork’s experimental Numba-oriented direction or are studying old internals. For production or current ML/accelerator work, the fork is too stale and diverged to be a safe default.

Prefer upstream unless you specifically need an old JAX snapshot; this fork is too stale to adopt for modern accelerator, compiler, or library work.

Prefer the upstream project unless you specifically need legacy JAX behavior or historical reproducibility. This fork is so far behind and so stale that it is not a good adoption target for modern accelerator or plugin work.