Skip to content

Stop overriding JAX GPU memory preallocation - #1539

Merged
t-kalinowski merged 1 commit into
mainfrom
agent/reduce-jax-gpu-preallocation
Jul 31, 2026
Merged

Stop overriding JAX GPU memory preallocation#1539
t-kalinowski merged 1 commit into
mainfrom
agent/reduce-jax-gpu-preallocation

Conversation

@t-kalinowski

@t-kalinowski t-kalinowski commented Jul 31, 2026

Copy link
Copy Markdown
Member

Summary

Stop setting XLA_PYTHON_CLIENT_MEM_FRACTION=1.00 when selecting the JAX GPU backend on Linux. This preserves either JAX's default allocation policy or a user-provided value.

Setting the fraction to 100% can produce a RESOURCE_EXHAUSTED message during CUDA initialization, even when JAX can recover and complete the operation.

Impact

Public-facing: use_backend("jax", gpu = TRUE) no longer overrides the JAX GPU memory allocation configuration.

Internal: no API or Python dependency changes.

Closes #1538

@t-kalinowski
t-kalinowski marked this pull request as ready for review July 31, 2026 11:51
@t-kalinowski
t-kalinowski merged commit 71d853a into main Jul 31, 2026
0 of 8 checks passed
@t-kalinowski
t-kalinowski deleted the agent/reduce-jax-gpu-preallocation branch July 31, 2026 12:24
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Example in Deep Learning with R, 3rd edition does not work with JAX backend on Linux / NVIDIA GPU

1 participant