mlx Files

What's Changed

• Preserve subnormal float values when casting to bool by @reckylurker in https://github.com/ml-explore/mlx/pull/4224
• Support assigning through a bare Ellipsis index by @Adityaj0 in https://github.com/ml-explore/mlx/pull/4314
• Fix divmod truncating the quotient for floats by @ayaangazali in https://github.com/ml-explore/mlx/pull/4108
• Add force_fused option to scaled_dot_product_attention by @hojin12312 in https://github.com/ml-explore/mlx/pull/4185
• Reject a negative eps in the normalization layers by @ayaangazali in https://github.com/ml-explore/mlx/pull/4312
• Bound GGUF metadata string/array values against the file mapping by @x14ngch3n in https://github.com/ml-explore/mlx/pull/4212
• Read each K/V byte once in gqa-8 decode attention by @dudududukim in https://github.com/ml-explore/mlx/pull/4077
• Fix fft vmap and jvp for transforms over a subset of axes by @kapellirohith in https://github.com/ml-explore/mlx/pull/4138
• Fix median dropping NaN by @devteamaegis in https://github.com/ml-explore/mlx/pull/4146
• Fix the CPU scan over a size one axis with a padded stride by @kapellirohith in https://github.com/ml-explore/mlx/pull/4139
• Validate the optimizer betas at construction by @ayaangazali in https://github.com/ml-explore/mlx/pull/4310
• RMSNormVJP backward writes a full {n_rows, D} gw_temp intermediate by @JasonHonKL in https://github.com/ml-explore/mlx/pull/4293
• [Bug]: add default none value to axis parameter of the take_along_axis by @aaishwarymishra in https://github.com/ml-explore/mlx/pull/4357
• Add a fused full-attention path for head_dim 256 on NAX devices by @wyanzhao in https://github.com/ml-explore/mlx/pull/3842
• Update nanobind to 2.15.0 by @XXXXRT666 in https://github.com/ml-explore/mlx/pull/4337
• Skip unnecessary simdgroup computations for quantised MOE matmuls on NAX by @RohanGautam in https://github.com/ml-explore/mlx/pull/4352
• Add AI usage policy by @zcbenz in https://github.com/ml-explore/mlx/pull/4331
• Raise cpu stream errors from synchronize by @robertomeroni in https://github.com/ml-explore/mlx/pull/4338
• chore: Validate eps in Adam at construction by @vraj00222 in https://github.com/ml-explore/mlx/pull/4361
• Bound winograd conv2d working set by tiling the batch by @Gusanidas in https://github.com/ml-explore/mlx/pull/4102
• Use a 32-row block in qmm_t_nax when one block covers all of M by @dwijenpatel in https://github.com/ml-explore/mlx/pull/4171
• Deduplicate fftshift and ifftshift by @Adityaj0 in https://github.com/ml-explore/mlx/pull/4318
• Fix Log and Equal is_equivalent ignoring primitive state by @kapellirohith in https://github.com/ml-explore/mlx/pull/4266
• Stabilize reduced-precision InstanceNorm by @ternaus in https://github.com/ml-explore/mlx/pull/4230
• Normalize negative axes in sort and argsort by @deBrian07 in https://github.com/ml-explore/mlx/pull/4332
• Clean up main thread compile cache before python interpreter shuts down by @zcbenz in https://github.com/ml-explore/mlx/pull/4373
• chore: Check malformed jaccl hostfile that miss rdma in pairs by @erwinzhang7 in https://github.com/ml-explore/mlx/pull/4284
• Round mxfp8 block scales up to avoid saturation by @dhiltgen in https://github.com/ml-explore/mlx/pull/4353
• Add support for the array_namespace_info by @aaishwarymishra in https://github.com/ml-explore/mlx/pull/4334
• Stop a failed CUDA graph commit from poisoning the encoder by @strayberry in https://github.com/ml-explore/mlx/pull/4356
• Fix quantized kernels in JIT build by @dwijenpatel in https://github.com/ml-explore/mlx/pull/4372
• [Metal][Performance] Avoid zero work in stride-2 ConvTranspose3d by @ternaus in https://github.com/ml-explore/mlx/pull/4343
• [CUDA] Ce fused kernel by @nastya236 in https://github.com/ml-explore/mlx/pull/3947
• Fix cpu exclusive scan for complex numbers by @ayaangazali in https://github.com/ml-explore/mlx/pull/4272
• Support Relocatable CUDA DLLs on Windows by @dhiltgen in https://github.com/ml-explore/mlx/pull/4382
• Use cast_to for fused AsType in compiled Metal kernels by @katlun-lgtm in https://github.com/ml-explore/mlx/pull/4351
• Declare DLPackCompatible protocol members as methods, not settable attributes by @Adityaj0 in https://github.com/ml-explore/mlx/pull/4384
• Fix quantizing sliced arrays by @zcbenz in https://github.com/ml-explore/mlx/pull/4381
• Fix einsum dropping a trailing empty subscript by @Adityaj0 in https://github.com/ml-explore/mlx/pull/4299
• Add script to run python tests by @zcbenz in https://github.com/ml-explore/mlx/pull/4393
• Hold GIL in AttachedData destructor by @zcbenz in https://github.com/ml-explore/mlx/pull/4391

New Contributors

• @hojin12312 made their first contribution in https://github.com/ml-explore/mlx/pull/4185
• @x14ngch3n made their first contribution in https://github.com/ml-explore/mlx/pull/4212
• @vraj00222 made their first contribution in https://github.com/ml-explore/mlx/pull/4361
• @ternaus made their first contribution in https://github.com/ml-explore/mlx/pull/4230
• @strayberry made their first contribution in https://github.com/ml-explore/mlx/pull/4356

Full Changelog: https://github.com/ml-explore/mlx/compare/v0.32.1...v0.32.2