Skip to content

Releases: microsoft/folx

v0.2.22

28 May 14:02

Choose a tag to compare

v0.2.21

28 May 10:50

Choose a tag to compare

What's Changed

  • Initial code commit by @jonkhler in #2
  • Create codeql.yml by @jonkhler in #4
  • Update codeql.yml by @jonkhler in #5
  • Delete .github/workflows/codeql.yml by @jonkhler in #6
  • Create publish.yaml by @jonkhler in #7
  • Update publish.yaml by @jonkhler in #9
  • update to version 0.2 by @n-gao in #10
  • Update README.md by @jonkhler in #8
  • added alternative operators by @n-gao in #11
  • fix detection of vectorization sequence by @n-gao in #12
  • support arbitrary input sizes for batched vmap by @n-gao in #14
  • support custom jvp; fix JHJ product; fix dtype conversion by @n-gao in #15
  • Fix zero Hessian entry removal for dense matrices by @n-gao in #16
  • Print source code lines when something goes wrong / add ruff by @n-gao in #17
  • Allow FwdLaplArray as input to wrapped functions by @n-gao in #18
  • Finalize support for complex numbers by @n-gao in #19
  • Use JAX's default rules for jnp.abs by @n-gao in #20
  • custom rules for complex abs by @n-gao in #21
  • Some fixes related to unmarked axis reduction and broadcasting of the sparsity mask by @n-gao in #22
  • add support for weighted inputs by @n-gao in #23
  • Add pallas fused attention and a corresponding forward Laplacian by @ae-foster in #26
  • avoid error when calling big_number in pallas kernel by @szbernat in #31
  • add pallas kernel for multi head self edge attention by @szbernat in #32
  • Use TritonCompilerParams in pallas_call by @ae-foster in #34
  • shard_map compatibility by @inailuig in #36
  • shard_map compatibility part 2 by @inailuig in #39
  • Handle the new JAX bind-params API by @gcarleo in #42
  • Fix jacrev/jacfwd varying basis construction for JAX 0.10 by @dalin27 in #43

New Contributors

Full Changelog: https://github.com/microsoft/folx/commits/v0.2.21