Repository Issue Activity (beta feature)

#

#3338 [BUG] mx.compile produces wrong results when chaining unsigned->signed astype with matmul

#3435 Missing precompiled Metal kernel for MoE gather_mm during LoRA/DoRA training (steel_gather_mm_rhs_nax_nt_float32_float32_bm64_bn128_bk128_wm2_wn4)

#3347 [BUG] mx.fast.metal_kernel: use-after-free when multiple custom kernels compose in lazy graph

#

#

#2708 [Feature Request] Complex Number and GPU Support for `mlx.linalg.svd` and `mlx.linalg.eig`

#

#

#3430 [BUG] mlx-cpu backend has a hard dependncy on a deprecated version of gcc

#

#

#3431 Feature Request: Expose underlying data pointer for mlx.core.array

#

#

#

#

#

#

#

#

#

#

#3402 MXFP4 quantization extremely slow on distributed GLM-5.1 (DeepSeek-V3.2 arch) — 0.27 tok/s vs 16 tok/s expected

#