v0.7.13
What's New
Added
- GLM model detection in
moe_mlppattern, with opt-in fused SwiGLU viaZMLX_GLM_FUSED_SWIGLUenv var - Experimental MoE stream pool:
ZMLX_MOE_STREAMS=Nfor multi-stream expert dispatch - Kernel correctness tests covering bits, quant, image, indexing, fused_moe, and optimizers
Fixed
- Numerically stable sigmoid:
kk_sigmoiduses abs+branch to avoid overflow on large negative inputs - SwiGLU native dtype: forward/backward kernels use native dtype with
kk_sigmoidhelper - mypy stream type errors in moe_mlp.py (DeviceType → Device)
Changed
- GLM auto-excluded:
moe_mlpandswiglu_mlpexcluded for GLM models (token fidelity failure) - README polish, gitignore cleanup
Full Changelog: v0.7.12...v0.7.13