Merge upstream MLX and preserve shared-buffer teardown safety - #11

Open

glg2672 wants to merge 89 commits into

Layr-Labs:mainfrom

glg2672:codex/merge/mlx-upstream

Commits on Jun 15, 2026

Bound Metal buffer COUNT, not just bytes, in MetalAllocator
anupsv
committed
d521341
Merge pull request #1 from Layr-Labs/fix/metal-resource-count-trim
anupsv
authored
008c51c

Commits on Jun 25, 2026

Bound Metal buffer COUNT, not just bytes, in MetalAllocator

anupsv
authored and
Gajesh2007
committed
aa480bd
Merge darkbloom-base: its resource-count trim is already included on top of 0.32.0 (this branch supersedes it)
Gajesh2007
committed
921d1ba
Merge pull request #3 from Layr-Labs/darkbloom/mlx-0.32.0-nax
Gajesh2007
authored
d5a2404

Commits on Aug 10, 2026

perf(mlx): opt-in Gemma 4 expert-QMM tile kernel with parallel descriptor builder (#4)
Gajesh2007
authored
a4b2b4c

Commits on Aug 13, 2026

Commits on Aug 14, 2026

Add endpoint parameter to linspace (#4184)

AKnassa
and
zcbenz
authored
1d717bd
Fix vmap of partition/argpartition dropping the kth argument (#4116)
Adityaj0
authored
306bdcd
Fix nan_to_num replacing inf with 0 for float16 and bfloat16 (#4222)

codeAnqiang-ma
and
zcbenz
authored
d9ad465
Fix einsum not broadcasting batch dimensions in batched tensordot (#4125)

Adityaj0
and
zcbenz
authored
bd5c3a2
Dequantize in float32 (#4241)
RohanGautam
authored
56e026d
chore: Reject complex in erf and erfinv (#4243)
ayaangazali
authored
6ba2d44
Fix cpu compilation failure of abs with uint (#4240)

ayaangazali
and
zcbenz
authored
3d23f7d
Fix quantize matrix multiplication floor issue (#4251)
JasonHonKL
authored
4947e3b
Only use MPI backend for world size > 1 (#4210)
erwinzhang7
authored
adf21de

Commits on Aug 15, 2026

chore: Reject complex in expm1, sigmoid and arctan2 (#4257)
ayaangazali
authored
140faa8
Decompose small kernel-depth 3D convs into 2D convs (#3785)

katlun-lgtm
and
zcbenz
authored
9ab977b

Commits on Aug 16, 2026

Commits on Aug 17, 2026

Remove grouped output in CI (#4195)
zcbenz
authored
61a4867
[CUDA] Fix finding cuda 13 headers in JIT compilation (#3995)
zcbenz
authored
0bb8121
Refactor wheel building script (#3818)
zcbenz
authored
d9e2b0d
Make mx.compile cache erasing thread safe (#4248)

zcbenz
and
yentur
authored
8e00a2d
Add builds for free-threaded python (#3812)
zcbenz
authored
9a79573
Fix int32 overflow in concatenate/repeat/kron (#4303)
ayaangazali
authored
98a188c
python: Widen list elements that do not fit in int32 to int64 (#4305)
ayaangazali
authored
e0fe540
Propagate CPU errors to events (#3742)

zcbenz
and
aleroot
authored
06f154b

Commits on Aug 18, 2026

Commits on Aug 19, 2026

Commits on Aug 20, 2026

Update nanobind to 2.15.0 (#4337)
XXXXRT666
authored
e2c1e28
Skip unnecessary simdgroup computations for quantised MOE matmuls on NAX (#4352)
RohanGautam
authored
c7ff35d
Add AI usage policy (#4331)

zcbenz
and
j-bowhay
authored
057cdc9
Raise cpu stream errors from synchronize (#4338)

robertomeroni
and
zcbenz
authored
994d9d5
chore: Validate eps in Adam at construction (#4361)

vraj00222
and
nastya236
authored
b3f1e1e
Bound winograd conv2d working set by tiling the batch (#4102)

Gusanidas
and
zcbenz
authored
27fec90

Commits on Aug 21, 2026

Commits on Aug 22, 2026

Commits on Aug 23, 2026

Round mxfp8 block scales up to avoid saturation (#4353)

authored
02adf7b
Add support for the __array_namespace_info__ (#4334)
aaishwarymishra
authored
7789905
Stop a failed CUDA graph commit from poisoning the encoder (#4356)

strayberry
and
zcbenz
authored
ab3ef95
Fix quantized kernels in JIT build (#4372)

dwijenpatel
and
zcbenz
authored
7408e68
Avoid zero work in stride-2 ConvTranspose3d (#4343)
ternaus
authored
d9077d8
[CUDA] Ce fused kernel (#3947)
nastya236
authored
451dc87

Commits on Aug 24, 2026

Fix cpu exclusive scan for complex numbers (#4272)

ayaangazali
and
zcbenz
authored
29b61c0
Support Relocatable CUDA DLLs on Windows (#4382)
dhiltgen
authored
9d173ff
Use cast_to for fused AsType in compiled Metal kernels (#4351)

authored
c793734
python: Declare DLPackCompatible protocol members as methods (#4384)
Adityaj0
authored
9d16475
Fix quantizing sliced arrays (#4381)
zcbenz
authored
a9eed5a
Fix einsum dropping a trailing empty subscript (#4299)

Adityaj0
and
zcbenz
authored
43d2f06
Merge remote-tracking branch 'upstream/main' into codex/merge/mlx-upstream
glg2672
committed
17a9316
fix: resolve upstream fp qvm merge
glg2672
committed
768b4c5
fix: remove duplicate qmv wide merge
glg2672
committed
1e93028

Commits on Aug 25, 2026

style: format merged MLX changes
glg2672
committed
bc42b3d
test: synchronize shared-buffer teardown
glg2672
committed
6d12ff2

Commits on Aug 26, 2026

Merge origin/main into codex/merge/mlx-upstream
glg2672
committed
2c000e5
fix: remove duplicate expert QMM device accessors
glg2672
committed
f647d24