Fsdp Python Packages

torchft-nightly

Fault tolerance for PyTorch (HSDP, LocalSGD, DiLoCo, Streaming DiLoCo)

133K 543 80

substrax

JAX/Flax NNX training infrastructure: device meshes and SPMD sharding, an Orbax checkpoint store, early stopping and callbacks, W&B and MLflow logging.

20K 0 0

torchft

Fault tolerance for PyTorch (HSDP, LocalSGD, DiLoCo, Streaming DiLoCo)

750 543 80

dewml

A nice ML framework based on Jax/Flax. Based on https://github.com/AshishKumar4/FlaxDiff

329 1 0

specforge

Train speculative decoding models effortlessly and port them smoothly to SGLang serving.

185 1K 364

dew-flash-attn-cu13

A nice ML framework based on Jax/Flax. Based on https://github.com/AshishKumar4/FlaxDiff

166 1 0

dew-flash-attn-cu12

A nice ML framework based on Jax/Flax. Based on https://github.com/AshishKumar4/FlaxDiff

151 1 0

tsugi-mend

Cross-rack distributed-training reducer for PyTorch. Apache-2.0, patent-independent. Part of the unified pip install tsugi surface.

128 1 1

tsugi

Unified developer surface for TsugiCinema's open-source distributed-training SDKs. pip install tsugi.

123 0 1

tsugi-kpool

K-Pool LoRA / Infinity software analog at LoRA adapter granularity. Apache-2.0, patent-aligned. Part of the unified pip install tsugi surface.

116 0 1