-
Currently, we are using deep copies of parameters in order to avoid a memory leak issue that resulted in MCMC chains overwriting an inappropriately shared value for a particular parameter. While this…
-
I was wondering if it would be possible to allow [`jax.pure_callback`](https://jax.readthedocs.io/en/latest/_autosummary/jax.pure_callback.html) to infer `result_shape_dtypes` by calling the specified…
-
We will need to simulate models from xspec since it is hard to port them all to pure JAX. Build a special class of SpectralModel that are evaluated through external softwares and in particular xspec.
-
I have been writing pallas kernels and has recently been made aware that they are not automatically partitioned.
I guess the general case is complicated; in my own case I am going with the `custom_…
luyug updated
9 months ago
-
### Description
Some background, I'm trying to parallelize a CPU-intensive computation using a callback to some `scipy.optimize` routines using `jax.pure_callback` across the available CPUs on my m…
-
From f2f discussion there are occasional stochastic failures in the NUTS sampling procedure. The characteristics of these failures are:
- They occur during NUTS warm up.
- They cause either errors w…
-
Is there any value in supplying a pure_callback (one with a return value), a buffer into which to place it's output? Something like the code below. I have some science applications where I need to use…
-
For example, there is a lookup table backend, and this operation need to get the table handle to reach it. This feature is implemented in tensorflow by tf.resource.
Can jax.pure_callback and jax.ex…
-
I found that for the common training pattern in Jax:
`new_state, other_output = jitted_train_step_fn(old_state, other_input)
`
Current XLA runtime may assign different backing device memory bu…
-
### Description
```bash
$ cd /tmp
$ git clone https://github.com/google/jax.git -b jax-v0.4.16
$ pip install matplotlib flax jaxlib tensorflow_datasets tensorflow_serving_api tensorflow -e jax
…