All issues

Search results

Concurrent import mlx.core across processes serialises (~0.2 s per worker); numpy stays flat

Status: Open.

#4268 In ml-explore/mlx;

· guillaume-osmo opened on Aug 15, 2026

quantized_matmul does not amortize the weight read for small M, while the bf16 path does

Status: Open.

#4265 In ml-explore/mlx;

· avlp12 opened on Aug 15, 2026

Would a fused residual-add + rms_norm be welcome in mx.fast?

question

Status: Open.

#4259 In ml-explore/mlx;

· Rahu378 opened on Aug 14, 2026

gather_qmm (sorted): ~69% of dense quantized_matmul throughput for MoE-typical small groups (~43 rows/expert, M3 Ultra)

Status: Open.

#4246 In ml-explore/mlx;

· avlp12 opened on Aug 14, 2026

Heap out-of-bounds write in GGUF quantized tensor loading: allocation size truncates to 32 bits

Status: Open.

#4244 In ml-explore/mlx;

· perparimmjeku opened on Aug 13, 2026

Add __array_namespace_info__() support to mlx.core

Status: Open.

#4239 In ml-explore/mlx;

· aaishwarymishra opened on Aug 13, 2026

[BUG] MLX python bindings reject int64 values like 2**40 in parameters.

bug

Status: Open.

#4238 In ml-explore/mlx;

· aaishwarymishra opened on Aug 13, 2026

InstanceNorm FP16 variance overflow collapses finite feature maps

Status: Open.

#4228 In ml-explore/mlx;

· ternaus opened on Aug 13, 2026

[BUG] mx.sort / topk / partition on Metal GPU return zeros for rows with negative strides

bug

Status: Open.

#4225 In ml-explore/mlx;

· reckylurker opened on Aug 13, 2026

Out-of-bounds read in GGUF metadata loader (unbounded string/array length)

Status: Open.

#4213 In ml-explore/mlx;

· x14ngch3n opened on Aug 12, 2026

[BUG] any/all and astype(bool) flush float subnormals to zero on Metal

bug

Status: Open.

#4205 In ml-explore/mlx;

· robertomeroni opened on Aug 12, 2026

qmm_splitk's 512-threadgroup target routes M<=32 transposed qmm to the non-NAX split-K kernel on M5 (M=33 is 21% faster than M=32)

Status: Open.

#4198 In ml-explore/mlx;

· pierre427 opened on Aug 12, 2026