-
### Description
I am experiencing hanging for jax versions newer than 0.4.31, as referenced in an earlier issue that I created and subsequently closed (https://github.com/jax-ml/jax/issues/24219). I …
-
I am encountering an index error when running [this](https://flax.readthedocs.io/en/v0.8.3/experimental/nnx/nnx_basics.html#:~:text=from%20functools%20import%20partial) example in the documentation
I…
-
### Please describe the purpose of the feature. Is it related to a problem?
Create a DreamerV3 implementation - there is no pure Jax implementation to date.
### Describe the solution you'd like
P…
-
### Description
When `vectorized=True` the expectation is that the `callback` of `pure_callback` should vectorise over common leading batch dims. That is, **all** batch dims of **any** mapped array s…
-
It might be good to have benchmarking comparisons for all unxt primitives and overrides against pure-numpy and pure-JAX to check for unexpected overheads in `jit`ted and `grad`ed functions that contai…
-
In example "execute_AD.py", replacing the lines
```
if __name__ == "__main__":
main(readjson(os.path.join(os.getcwd(), "settings_AD.json")))
```
by
```
if __name__ == "__main__":
se…
-
I would like to be able to write a callback that accepts accepts unbatched arguments, which would normally be executed under `pure_callball(... vmap_method='sequential')`, but then perform `vmap` usin…
-
### 🚀 The feature, motivation and pitch
Pitch: Get the forward/backward of a Dynamo function
We (PyTorch/XLA) would like to get a symbolic graph representation (e.g. FX graph) of the forward and b…
-
Laundry list of things missing in the Metal backend when I tried using it:
- jax.numpy.linalg.lstsq
`NotImplementedError: MLIR translation rule for primitive 'eigh' not found for platform METAL`…
-
Hi team,
First off, I absolutely love JAX. It's the core engine behind our startup.
It would be fantastic to have a rank-one update to an eigenvalue decomposition of a symmetric PSD matix $A$. I…