Repository navigation

All pull requests

Fix vmap of tri_inv ignoring the triangular flags
#4748mahimn01 opened Oct 22, 2026
Fixes MLX/ErrorHandler.swift:345: Fatal error due to missing stream argument
#4746Satchitananda opened Oct 22, 20261 of 2 tasks
[CUDA] Avoid multi-root graphs on DGX/RTX spark
#4745dhiltgen opened Oct 22, 2026Contributor
1
Fix GPU pad for outputs larger than 2^31 elements
#4744tivris opened Oct 22, 2026
Use gemv_wide for fp32 matmuls with 2-3 rows
#4742utsav1033 opened Oct 22, 2026
Cleanup metal quantized kernel instantiations
#4741zcbenz opened Oct 22, 2026Member
Fix FFT read/write past buffer end in partial threadgroups
#4740kinetrix opened Oct 21, 2026
Fix CUDA gather_qmm for experts past 2^32 weight elements
#4738IamCooller opened Oct 21, 2026
[ring] Only touch the socket task queues while holding the lock
#4737Shakhtar-Sankur opened Oct 21, 20261 task done
Fix/padding jvp vjp
#4736michalk8 opened Oct 21, 2026Collaborator
Adding sliding window attention
#4734jagrit06 opened Oct 20, 2026Member
3
feat(nn): add freqs and partial_rotary_factor support to RoPE
#4732shadyuwugurl opened Oct 20, 2026
Fix zero-copy numpy arrays dtype mismatch
#4731michalk8 opened Oct 20, 2026Collaborator
Fix normal loc scale shape
#4730michalk8 opened Oct 20, 2026Collaborator
Fix mx.take's negative indices
#4729michalk8 opened Oct 20, 2026Collaborator
Widen affine qmv_wide row tiles to 8
#4728Ambisphaeric opened Oct 20, 20261 task done
1
Use data_size for Metal gather/scatter large-offset checks
#4724begcdn opened Oct 19, 2026
[Metal] Add SDPA fast paths for speculative decoding
#4723dhiltgen opened Oct 18, 2026Contributor
Add optional int8 path for quantized matmul with affine weights
#4722magnuslundstedt opened Oct 18, 2026Contributor
1
Fix silent Metal grid truncation in softmax, logsumexp, rms_norm and layer_norm past 2^27 rows
#4717ArupKantiDas opened Oct 17, 2026
1
Fix sharded-to-all bias and integer split positions on quantized layers
#4715ronaldmannak opened Oct 17, 2026Contributor
Debug time graph verification
#4711zcbenz opened Oct 16, 2026Member
Fix Windows CUDA 13 DLL loading
#4706luciusacarr opened Oct 15, 2026
Fix fftn size one axis aliasing a temporary
#4705StockerMC opened Oct 15, 2026
2
Make GPU scheduler know the spin is waiting for something
#4703aleroot opened Oct 15, 2026Contributor
3

Navigation Menu

Repository navigation

All pull requests

Fix vmap of tri_inv ignoring the triangular flags

Fixes MLX/ErrorHandler.swift:345: Fatal error due to missing stream argument

[CUDA] Avoid multi-root graphs on DGX/RTX spark

Fix GPU pad for outputs larger than 2^31 elements

Use gemv_wide for fp32 matmuls with 2-3 rows

Cleanup metal quantized kernel instantiations

Fix FFT read/write past buffer end in partial threadgroups

Fix CUDA gather_qmm for experts past 2^32 weight elements

[ring] Only touch the socket task queues while holding the lock

Fix/padding jvp vjp

Adding sliding window attention

feat(nn): add freqs and partial_rotary_factor support to RoPE

Fix zero-copy numpy arrays dtype mismatch

Fix normal loc scale shape

Fix `mx.take`'s negative indices

Widen affine qmv_wide row tiles to 8

Use data_size for Metal gather/scatter large-offset checks

[Metal] Add SDPA fast paths for speculative decoding

Add optional int8 path for quantized matmul with affine weights

Fix silent Metal grid truncation in softmax, logsumexp, rms_norm and layer_norm past 2^27 rows

Fix sharded-to-all bias and integer split positions on quantized layers

Debug time graph verification

Fix Windows CUDA 13 DLL loading

Fix fftn size one axis aliasing a temporary

Make GPU scheduler know the spin is waiting for something

Pull requests