Registry / ai-ml / jax-cuda12-pjrt

jax-cuda12-pjrt

JSON →
library0.11.1pypypi✓ verified 22d ago

The jax-cuda12-pjrt package provides the JAX XLA PJRT backend for NVIDIA GPUs, specifically built with CUDA 12. It serves as the `jaxlib` implementation when GPU acceleration is desired. The current version is 0.9.2, and JAX along with its ecosystem components typically follow a rapid release cadence, often with monthly or bi-monthly updates.

pip install jax jax-cuda12-pjrt
INSTALL
IMPORT
SIG · JAX-CUDA12-PJRT
J
jax-cuda12-pjrt
ai-mlpythonv0.11.1
Install
16.3s avg
Import
1940ms
Disk
1106MB
Pass rate
5/ 10
Env Coverage5 / 10
glibc
3.93.13
musl
3.93.13
Install & Compatibility
Where this runs
tested against v0.11.1 · pip install
no network on importno background threads
Install × environment matrix
Each cell = how many times install + import succeeded across repeated harness runs. Partial = flaky.
glibc = Debian/Ubuntu slim · musl = Alpine Linux
musl
py 3.103.95 runs
build_error
glibc
py 3.103.95 runs
installs and imports cleanly · install 16.3s · import 1.940s · 1126.4MB
1106MB installed
● package 1106MB
Code
Verified usage

Verified import paths — ran on the pinned version, not inferred.

jax
import jax
import jax_cuda12_pjrt
The `jax-cuda12-pjrt` package provides the backend implementation (jaxlib) and is not directly imported into user code. All interactions are through the core `jax` library.

This quickstart demonstrates basic JAX usage. It checks the JAX backend and available devices, then defines and executes a JIT-compiled function, confirming that the computation leverages GPU acceleration if available.

import jax import jax.numpy as jnp # Check for available devices print(f"JAX backend: {jax.default_backend()}") print(f"Available devices: {jax.devices()}") # Define a JIT-compiled function @jax.jit def sum_of_squares(x): return jnp.sum(x**2) # Create some data data = jnp.array([1.0, 2.0, 3.0, 4.0]) # Run the function result = sum_of_squares(data) print(f"Input data: {data}") print(f"Result (sum of squares): {result}") # Verify it's on a device if available if jax.devices('gpu'): print(f"Result device: {result.device()}")
Debug
Known issues
breakingThe default `jax.pmap` implementation has changed, and `jax.pmap` is now in maintenance mode. Users are encouraged to migrate to `jax.shard_map` for new code and distributed computations.
fix
Refer to the JAX migration guide for `pmap` (docs.jax.dev/en/latest/migrate_pmap.html) and use `jax.shard_map` instead of `jax.pmap`.
affects: >=0.8.0
breaking`jax.dlpack.from_dlpack` no longer accepts a DLPack capsule directly. It now requires an array implementing `__dlpack__` and `__dlpack_device__`.
fix
Update code to pass an array-like object that implements the DLPack protocol, rather than a raw capsule, to `jax.dlpack.from_dlpack`.
affects: >=0.7.2
deprecated`jax.lax.pvary` has been deprecated.
fix
Use `jax.lax.pcast(..., to='varying')` as the replacement for `jax.lax.pvary`.
affects: >=0.8.2
gotcha`jax.numpy.arange` with a `step` argument no longer generates the array on the host by default. This change improves efficiency but can lead to less precise outputs for narrow-width floats (e.g., bfloat16).
fix
To recover the previous host-based generation and ensure higher precision for narrow-width floats, explicitly cast the NumPy output: `jnp.array(np.arange(...))`.
affects: >=0.9.2
gotchaWhen using `jax.shard_map` in 'Explicit' mode, JAX will now raise an error if the `PartitionSpec` of an input does not match the `PartitionSpec` specified in `in_specs`. Previously, this might have silently caused an implicit reshard.
fix
Ensure that the `PartitionSpec` of your inputs precisely matches the `in_specs` argument when using `jax.shard_map` in 'Explicit' mode. Omit `in_specs` if you intend for `shard_map` to infer the partitioning.
affects: >=0.9.1
gotchaInstalling `jaxlib` (the CPU-only version) alongside a GPU-specific `jaxlib` variant like `jax-cuda12-pjrt` can lead to conflicts, unexpected device selection, or errors. Only one `jaxlib` implementation should be installed.
fix
Ensure only `jax` and your desired `jaxlib` variant (e.g., `jax-cuda12-pjrt` for CUDA 12 GPU or `jaxlib` for CPU) are installed. Uninstall any conflicting `jaxlib` packages before installing your target GPU backend.
affects: All versions
Upgrade
Version history
0.11.1latest on PyPI · released Aug 17, 2026
Audit
Dependencies
jaxrequiredThis package provides the GPU backend for the core JAX library. JAX itself is required for usage.
Agent activity
7 hits · last 30 days
node
6
Resources
jax-cuda12-pjrt — pip install jax-cuda12-pjrt · libregistry