Default Branch

382808c14b · ci : re-enable rocm build on amd64 (#18439) · Updated 2025-12-28 15:29:23 -08:00

Branches

3c8a2a83fe · shmem experiments · Updated 2024-11-26 05:17:38 -08:00

3386
3

dafedd33d2 · 4x4 -> 4x · Updated 2024-11-26 04:54:02 -08:00

3386
2

bf3494345e · metal : some mul_mv experiments · Updated 2024-11-26 04:48:50 -08:00

3386
1

b83cae088c · speculative : add infill mode · Updated 2024-11-26 01:14:17 -08:00

3391
1

4ff0831ce6 · metal : use F16 math in mul_mat kernels · Updated 2024-11-25 05:15:26 -08:00

3404
1

f7b0233eca · wip · Updated 2024-11-16 00:33:55 -08:00

3466
1

5e6dad9322 · speculative : experimenting with Qwen2.5 · Updated 2024-11-14 01:31:31 -08:00

3488
2

33bdee667e · speculative : fix out-of-bounds access · Updated 2024-11-14 01:23:45 -08:00

3488
1

8c1b186cb5 · metal : minor Q4_0 optimization · Updated 2024-11-12 05:30:51 -08:00

3498
21

3d1fe1bb4d · metal : int -> short, style · Updated 2024-11-09 00:32:16 -08:00

3509
2

bd1198a67a · metal : fix build and some more comments · Updated 2024-11-09 00:09:50 -08:00

3509
1

a2385da59c · make : clean-up [no ci] · Updated 2024-11-08 03:46:20 -08:00

3516
9

94accca4c2 · vec move mask to shmem · Updated 2024-11-07 10:58:10 -08:00

3526
19

c5d8bb5a81 · leave only basic functions for SYCL CI · Updated 2024-11-05 23:47:50 -08:00

3591
2

4fc8673d09 · llama-bench : skip repeated values in consecutive lines · Updated 2024-11-02 07:37:33 -07:00

3551
1

20e12112fd · llama : suggest reduce ctx size when kv init fails · Updated 2024-11-01 16:55:19 -07:00

3554
2

afc4a7de65 · llama : enable flash attn automatically when supported · Updated 2024-10-30 15:30:06 -07:00

3571
1

8233009d4d · Support SYCL device register · Updated 2024-10-19 19:06:51 -07:00

3652
1

bc82fc2ed8 · llama-bench : add time-to-first-byte stat · Updated 2024-10-18 06:40:02 -07:00

3623
1

2d3fc54ac6 · add amx kernel for gemm · Updated 2024-10-17 20:35:49 -07:00

3633
1