Interactive causes bazel to output more
useful info when running locally.
* Fix issue with rocm el8 repo urls
Work around quirk with rocm version
when it ends with 0
* Fix package name conflict
Ubu22 and higher have a package name conflict
between the debian versions and the AMD provided
versions.
* [ROCm] Use clang env
1) Hermetic CUDA rules allow building wheels with GPU support on a machine without GPUs, as well as running Bazel GPU tests on a machine with only GPUs and NVIDIA driver installed. When `--config=cuda` is provided in Bazel options, Bazel will download CUDA, CUDNN and NCCL redistributions in the cache, and use them during build and test phases.
[Default location of CUNN redistributions](https://developer.download.nvidia.com/compute/cudnn/redist/)
[Default location of CUDA redistributions](https://developer.download.nvidia.com/compute/cuda/redist/)
[Default location of NCCL redistributions](https://pypi.org/project/nvidia-nccl-cu12/#history)
2) To include hermetic CUDA rules in your project, add the following in the WORKSPACE of the downstream project dependent on XLA.
Note: use `@local_tsl` instead of `@tsl` in Tensorflow project.
```
load(
"@tsl//third_party/gpus/cuda/hermetic:cuda_json_init_repository.bzl",
"cuda_json_init_repository",
)
cuda_json_init_repository()
load(
"@cuda_redist_json//:distributions.bzl",
"CUDA_REDISTRIBUTIONS",
"CUDNN_REDISTRIBUTIONS",
)
load(
"@tsl//third_party/gpus/cuda/hermetic:cuda_redist_init_repositories.bzl",
"cuda_redist_init_repositories",
"cudnn_redist_init_repository",
)
cuda_redist_init_repositories(
cuda_redistributions = CUDA_REDISTRIBUTIONS,
)
cudnn_redist_init_repository(
cudnn_redistributions = CUDNN_REDISTRIBUTIONS,
)
load(
"@tsl//third_party/gpus/cuda/hermetic:cuda_configure.bzl",
"cuda_configure",
)
cuda_configure(name = "local_config_cuda")
load(
"@tsl//third_party/nccl/hermetic:nccl_redist_init_repository.bzl",
"nccl_redist_init_repository",
)
nccl_redist_init_repository()
load(
"@tsl//third_party/nccl/hermetic:nccl_configure.bzl",
"nccl_configure",
)
nccl_configure(name = "local_config_nccl")
```
PiperOrigin-RevId: 662981325
Use get_rocm.py changes in ci_build to pull in
development builds for ROCm.
Specify ROCM_BUILD_JOB and ROCM_BUILD_NUM for
activating the development build path.
When `build_gpu_plugin` is true, three wheels will be produced (jaxlib, jax-cuda-pjrt and jax-cuda-plugin). If they are editable, they need to be placed in subdirectories to avoid overwrite.
Tested on GPU. After the editable wheels are built, they can be installed with `pip install -e /jax/dist/jax_gpu_pjrt /jax/dist/jaxlib /jax/dist/jax_gpu_plugin`.
PiperOrigin-RevId: 660984311
The plugin is released and the flag is no longer needed.
Also set default value of enable_gpu to False. enable_gpu will be removed in the next change.
PiperOrigin-RevId: 660059432
This lets us avoid bundling a whole another copy of LLVM with JAX packages
and so we can finally start building Mosaic GPU by default.
PiperOrigin-RevId: 638569750
At the moment we can export a JAX function into an Exported and we can invoke an Exported from another JAX function, but there is no way to serialize an Exported to be able to use it in another process.
Exported now has all the features we had in mind, so it is a reasonable time to add a serialization method. The intention is for the serialization to have backwards compatibility guarantees, meaning that we can deserialize an Exported that has been serialized with an older version of JAX. This PR does not add explicit support for versioning, nor backwards compatibility tests. Those will follow.
Here we add serialization and deserialization to bytearray, using the flatbuffers package. We use flatbuffers because it is simple, it has backwards and forwards compatibility guarantees, it is a lightweight dependency that does not require additional build steps, and it is fast (deserialization simply indexes into the bytearray rather than creating a Python structure).
In the process of implementing this we have done some small cleanup of the Exported structure:
* renamed serialization_version to mlir_module_serialization_version
* renamed disabled_checks to disabled_safety_checks
This code is tested by changing export_test.py to interpose a serialization followed by a deserialization every time we export.export.
There is a known bug with the serialization of effects, so I disabled one of the export tests. Will fix in a subsequent PR.
PiperOrigin-RevId: 590078785
