Paulo Valente

Add support for load balancing

2

The gRPC.io blog has a [great post](https://grpc.io/blog/grpc-load-balancing/) regarding load balancing. Since some use cases for gRPC don't have an explicit proxy between services, we should investigate how to implement it...

Add Mint client adapter

11

After #199, we now have a contract defined for the HTTP client, so we can now implement an adapter using the Mint library and use that as a default instead...

feat: recursion support

1

Improves the ideas from #61 for handling more edge cases

This PR aims to add an option so the users can force some fields to always be rendered as plain strings. This is useful for when indexing engines suffer with...

XlaCallModule or another way of loading StableHLO code

4

## 🌱 Describe your Feature Request Are there any plans on supporting a venue for loading StableHLO code? Currently, I can generate a StableHLO MLIR module and load it into...

Dependency mismatch between cowboy and gun 2.0

4

Currently, `gun` 2.0 depends on `cowlib` 2.12.0, but `cowboy` 2.9 depends on `cowlib` 2.11. This makes it impossible for `gun` to be updated to 2.0 on an Elixir project that...

Re-implement Nx.LinAlg.eigh as defn

6

Currently, we have a custom implementation for Nx.BinaryBackend and call the XLA implementation for eigh in EXLA. However, the XLA implementation seems to suffer from similar issues to the SVD...

Convert Nx.LinAlg.lu to optional callback

2

XLA doesn't provide an implementation, and we use LU decomposition for `determinant`, which is used in some other LinAlg functions. Therefore, we should either look into JAX's implementation or just...

Passing f32 data into LSTM with Axon.Loop trainer+run causes while shape mismatch error

8

```elixir input = Axon.input("input_series", shape: put_elem(Nx.shape(time_x), 0, nil)) model = input |> Axon.lstm(128, activation: :relu) |> elem(0) |> Axon.dropout(rate: 0.25) |> Axon.lstm(64, activation: :relu) |> elem(0) |> Axon.dropout(rate: 0.25) |>...