Trylle

Loading

Search Explore Sign upCreate an account Sign In

Commits · goniz/mlx | Trylle

gonizgoniz/mlxmlx

Public

Pull mirror status: ok

Code Issues 0 Releases 0 Actions

Commits on Mon, Aug 24, 2026

Goni Zahavy
Merge pull request #75 from goniz/cursor/spec-validation-tighten
4527d69Aug 24
Goni Zahavy
fix(vulkan): tighten matmul/FA env envelope checks
a75153eAug 24
Goni Zahavy
Merge pull request #73 from IlyaasK/fix/vulkan-spec-validation-guards
0a6a988Aug 24
Goni Zahavy
Merge pull request #74 from goniz/cursor/decode-batch-cleanup
70974aaAug 24
Goni Zahavy
chore(vulkan): tidy Apple decode-batch defaults
4680e30Aug 24
Goni Zahavy
Merge pull request #72 from IlyaasK/feat/agx-decode-batching
5dcb849Aug 24
Goni Zahavy
Merge pull request #71 from IlyaasK/feat/agx-flash-attn-tuning
66bee72Aug 24
Goni Zahavy
Merge pull request #70 from IlyaasK/feat/agx-scalar-mul-mm-tuning
e9bcfceAug 24
Goni Zahavy
Merge pull request #69 from IlyaasK/fix/asahi-apple-gpu-classification
d948c99Aug 24

Commits on Sun, Aug 23, 2026

IlyaasK
fix(vulkan): reject tuning specs outside numerically validated envelopes
3ed0900Aug 23
IlyaasK
perf(vulkan): default long decode recordings on Apple AGX
6441d51Aug 23
IlyaasK
perf(vulkan): tune scalar flash-attention params for Apple AGX
bdbac5aAug 23
IlyaasK
perf(vulkan): tune scalar mul_mm tiles for Apple AGX (BK=4, TM=4)
b2fc270Aug 23
IlyaasK
fix(vulkan): recognize Mesa honeykrisp Apple GPUs via device-name fallback
542a868Aug 23

Commits on Sat, Aug 1, 2026

Goni Zahavy
Merge pull request #68 from goniz/laguna
e806d94Aug 1

Commits on Thu, Jul 23, 2026

Goni Zahavy
[Vulkan] Guard zero dims in NVFP4 gather batch divides
1b5afc4Jul 23

Commits on Wed, Jul 22, 2026

Goni Zahavy
[Vulkan] Reuse NVFP4 weights across sorted MoE pairs
a8e04f0Jul 22
Goni Zahavy
[Vulkan] Speed up NVFP4 matvec with pack-16 K stepping
fb87d52Jul 22

Commits on Tue, Jul 21, 2026

Goni Zahavy
[Vulkan] Use BN=1 NVFP4 matvec for small-batch decode
1e9f56bJul 21
Goni Zahavy
[Vulkan] Multi-column NVFP4 matvec for MoE prefill
a494765Jul 21
Goni Zahavy
[Vulkan] Speed up NVFP4 prefill with register-blocked tiles
247199aJul 21
Goni Zahavy
[Vulkan] Add dense NVFP4 matvec/mm with bf16 activations
965bc22Jul 21
Goni Zahavy
[Vulkan] Add NVFP4 gather matvec for MoE decode
4d09707Jul 21
Goni Zahavy
[Vulkan] Fuse NVFP4 GatherQMM and spill UMA allocs to host heap
7f13bcbJul 21

Commits on Thu, Jul 16, 2026

Goni Zahavy
Merge pull request #67 from goniz/cursor/gdn-subgroup-clustered-cap-a20e
ea5289cJul 16
Goni Zahavy
[Vulkan] Expose subgroup_clustered in device_info
211fc07Jul 16
Goni Zahavy
Merge pull request #66 from goniz/cursor/qwen36-27b-dense-affine4-2578
430a327Jul 16
Goni Zahavy
[Vulkan] Fall back SDPA VJP to composed graph
05644d6Jul 16
Goni Zahavy
[Vulkan] Cache F16-promoted B^T for BF16 coopmat1 GEMMs
6ae98f9Jul 16
Goni Zahavy
[Vulkan] Harden coopmat aligned/K and coop4 subgroup safety
e6b880bJul 16