-
### Description
Hi, we encounter an `XlaRuntimeError: INTERNAL: unsupported operands type` that is hard to understand and debug.
This started to happen on some unit tests that we have on our CI:
…
-
The conflict is caused by:
praxis 1.4.0 depends on tfds-nightly==4.8.3.dev202303280045
The user requested jax==0.4.26
jax[cuda12] 0.4.26 depends on jax 0.4.26 (from https://pypi.tuna.ts…
-
Note the redundancy and all the "Alias for field number X" in classes such as optax.MultiStepsState: https://optax.readthedocs.io/en/latest/api/optimizer_wrappers.html#optax.MultiStepsState
![image…
-
Hi, thank you very much for providing such brilliant work and code, which greatly facilitates my research. I would like to inquire about the specific versions of the packages listed in the requirement…
-
Hello!
When it comes to annotations, `optax` currently relies heavily on `optax.Updates` and `optax.Params`, which are all aliases for `chex.ArrayTree`.
This makes sense, but for folks who run t…
-
Optax Throws attribute error when using the attached Google Colab Inference Demo
```python
---------------------------------------------------------------------------
AttributeError …
-
Hi,
I have my RAM getting used up to overflow when I use `scale_by_radam` gradient transform or equivalently `optax.radam` **without** JIT compiling the code. The problem appears on CPU and GPU but…
-
Dear All-
I have a very simple question. I have two neural networks of type `MLP` and I want to initialize optimizer via `optax`.
When I have one neural network I do like
`pyt1 = eqx.filter(bn, e…
-
I was trying to run Vivit model according to the _quick start_, I obtained this error:
**TypeError: get_optimizer() missing 1 required positional argument: 'learning_rate_fn'**
Since the train_lib_…
-
The [example in the docs](https://nemos.readthedocs.io/en/latest/generated/api_guide/plot_05_batch_glm/) currently uses a custom loop to implement stochastic gradient descent.
An alternative would …