Comments (3)
I think this bug has already been fixed at head, but it made it into the last release because we don't have any way to test SLURM clusters in CI.
Can you try with a nightly jaxlib and jax?
from jax.
You are right, it is indeed already fixed on the nightly build.
It works as expected using
jax==0.4.27.dev20240410
jaxlib==0.4.27.dev20240410+cuda12.cudnn89
Thanks!
from jax.
Ran into the same issue without using SLURM, but multi-node GPUs.
Same fix worked with these too:
jax==0.4.27.dev20240415
jaxlib==0.4.27.dev20240415+cuda12.cudnn89
Just dropping the install lines here in case anyone needed it like I did (got it from MaxText)
pip install --pre -U jax -f https://storage.googleapis.com/jax-releases/jax_nightly_releases.html
pip install -U --pre jaxlib -f https://storage.googleapis.com/jax-releases/jaxlib_nightly_cuda12_releases.html
# Maybe Transformer engine for H100+?
pip install "transformer-engine==1.5.0+297459b" \
--extra-index-url https://us-python.pkg.dev/gce-ai-infra/maxtext-build-support-packages/simple/
from jax.
Related Issues (20)
- Dot product results differ after sharding HOT 2
- jax.config has no attribute 'update' HOT 2
- Move jax.nn.log_softmax to jax.scipy.special? HOT 3
- Provide output buffer for pure_callback result HOT 3
- key reuse detection fails for closed-over keys in JIT
- Automatic sharded parallelization of per-sample gradients fails on GPUs HOT 2
- [v0.4.26] Fails to initialize CUDA backend after jax_cuda_visible_devices is set HOT 1
- [README] nn libraries fix HOT 26
- jax.lax.scan() segmentation fault with jax-metal on Mac M1 HOT 1
- Latency Hiding Scheduler leads to x5 memory usage if used without jax.lax.scan
- Wrong output of `jax.scipy.special.sph_harm` HOT 3
- Print large value with suppress in printoptions HOT 1
- An uninformative error on TPU HOT 2
- Make custom VJP bwd shape/type checking optional.
- Probability density/mass functions from jax.scipy.stats not supported on Metal
- Build fails due to /usr/bin/ccache being detected as Clang by build/build.py
- jnp.select should use jax.lax.select_n
- Let `xs=None` by default in `lax.scan` HOT 1
- jax.nn.softmax is inconsistent under jax.jit HOT 7
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from jax.