Karan Shah

Tutorials/Examples under a different namespace

2

To ensure velocity of changes/features in mainline, updating examples in parallel to comply with the release would consume time. Suggestion is to transition tutorials/examples under `openfl-examples` repository, which has its...

Added a prerequisite step to upgrade pip before installing opencv-python. This is the error while building container **without** the fix: ``` Collecting opencv-python Downloading https://files.pythonhosted.org/packages/77/f5/49f034f8d109efcf9b7e98fbc051878b83b2f02a1c73f92bbd37f317288e/opencv-python-4.4.0.42.tar.gz (88.9MB) Complete output from command...

Unknown term in `sample_permutation` for sinkhorn

1

Hi, I am trying to use `sinkhorn` matcher for batch sizes >8 (per device) as in the code. It (understandably) fails at #L62 since `bs * dim` exceeds sampling range...

Refactored DETR implementation on `optax` API.

2

Many functions and calls within existing `detr` implementation are deprecated in Flax as well as OTT. This PR includes two major changes (all changes only under `projects/baselines/detr`): * Migrates from...

Suboptimal default initialization of q/k/v projections in `nn.MultiHeadDotProductAttention`

2

Initialization of q/k/v projections are not forward-backward normalized for [`linen.MultiHeadDotProductAttention`](https://flax.readthedocs.io/en/v0.8.0/_modules/flax/linen/attention.html#MultiHeadDotProductAttention). This implementation does not face optimization issues when a `pre-LN` variant of transformer is used; but faces convergence issues in...

These are not used throughout the framework. `TensorDB.get_tensor_from_cache` is the recommended method to use.

Implementation of DETR using JAX/Flax.

CUDA `XlaRuntimeError` with MPI on `jax==0.4.31`

5

### Description Hi, `jax.jit` on a function seems to fail when running in an OpenMPI environment. An MWE is shown below: ```python # error.py # Run as: mpirun -n 8...

Hi, I think the following typecast is potentially unnecessary or with a wrong type. https://github.com/google/flax/blob/d718655f62edbb1c4227281f158b94b78e886814/flax/linen/linear.py#L1123-L1125 * First, in LLMs where `num_embeddings` is of the order 50-100K rows, `__call__` is generally...

[openfl-docker] Remove startup script dependency

1

This PR removes dependency on actor script to start demo federation, for readability. No documentation changes are expected as a result. Test(s): - [x] Dockerization CI