Skip to content

0.10.6

Latest
Latest
Compare
Choose a tag to compare
@cgarciae cgarciae released this 23 Apr 20:26
· 25 commits to main since this release
v0.10.6
0fde636

What's Changed

  • Sow top activations based on absolute value. by @copybara-service in #4670
  • Add support for layer-specific rope scale factors. by @copybara-service in #4672
  • Automatic model selection for Gemma 3 models. by @copybara-service in #4671
  • Make LoRA's dtype arg useful by @IvyZX in #4681
  • [NVIDIA] Support FP8 Einsum Op by @kaixih in #4686
  • [nnx] remove deprecated APIs by @cgarciae in #4627
  • Add attention_bias parameter to MultiHeadDotProductAttention. by @copybara-service in #4694
  • Unit tests for attention_bias parameter to MultiHeadDotProductAttention. Add parameter to all overloads to make pytype happy. by @copybara-service in #4702
  • Rollback of attention_bias parameter, because the change overrides the attention bias for injected attention functions. by @copybara-service in #4703
  • Add custom einsum op to Einsum() by @IvyZX in #4705
  • [nnx] refactor GraphDef by @cgarciae in #4630
  • Make fully replicated array before saving checkpoints for examples that use pmap. by @copybara-service in #4707
  • Fix CI by @cgarciae in #4716
  • remove "nnx" collection in ToLinen by @copybara-service in #4708
  • [nnx] flaxlib types by @cgarciae in #4639
  • v0.10.6 by @cgarciae in #4724

Full Changelog: v0.10.5...v0.10.6

Contributors

kaixih, cgarciae, and IvyZX
Assets 2
Loading