Default Branch

846e991ec3 · cuda: add sqrt_softplus in topk-moe for dsv4 (#25896) · Updated 2026-07-21 16:30:01 +00:00

Branches

93fbd407f3 · Merge branch 'master' into compilade/convert-prequant · Updated 2025-10-23 18:23:12 +00:00    furyhawk

3255
6

f0076dc5a0 · metal : adjust .get_alloc_size to be alloc friendly · Updated 2025-10-19 14:20:54 +00:00    furyhawk

3285
1

96f9f391c7 · ggml : fix unaligned access in AMX code · Updated 2025-09-29 07:37:15 +00:00    furyhawk

3465
1

a8b0089a5b · ggml : remove SVE paths · Updated 2025-09-28 17:26:03 +00:00    furyhawk

3465
1

837b1b4563 · ggml : remove KQ mask padding · Updated 2025-09-28 15:10:17 +00:00    furyhawk

3468
6

17ca6ed540 · Implement llama-pull tool · Updated 2025-09-20 16:25:21 +00:00    furyhawk

3556
1

e83ef74733 · one less magic number · Updated 2025-09-20 05:58:36 +00:00    furyhawk

3575
6

652d303b32 · metal : fuse add + rms · Updated 2025-09-18 13:29:25 +00:00    furyhawk

3573
1

64c6dcbe6d · metal : make the NSG a function constant in mul_mv kernels · Updated 2025-09-18 08:31:59 +00:00    furyhawk

3578
2

6045c5a263 · cont : put all buffers in the same virtual address space · Updated 2025-09-14 12:46:57 +00:00    furyhawk

3614
2

3f62ee8bee · metal : back to a single queue per device · Updated 2025-09-09 14:06:46 +00:00    furyhawk

3656
9

3f62ee8bee · metal : back to a single queue per device · Updated 2025-09-09 14:06:46 +00:00    furyhawk

3656
9

7b717fb4b2 · Rewrite llama-run to use llama-server · Updated 2025-09-05 16:22:36 +00:00    furyhawk

3693
1

9f2636b7dc · wip · Updated 2025-09-01 08:17:56 +00:00    furyhawk

3742
1

4317d5abf5 · wip · Updated 2025-08-28 10:55:21 +00:00    furyhawk

3776
1

dc2187d48d · ggml : fix SSM_SCAN for n_groups > 1 · Updated 2025-08-27 21:37:04 +00:00    furyhawk

3781
1

fb573f4440 · ggml-quants : avoid division by zero in make_q3_quants · Updated 2025-08-17 22:26:02 +00:00    furyhawk

3896
2

220860aa0c · graph : use F32 accumulators for gpt-oss · Updated 2025-08-14 13:08:31 +00:00    furyhawk

3923
1

d9b625edb6 · ggml-quants : handle imatrix for MXFP4 · Updated 2025-08-12 02:12:10 +00:00    furyhawk

3949
1

2763dc8b53 · ggml-quants : handle zero amax for MXFP4 · Updated 2025-08-06 20:26:25 +00:00    furyhawk

3987
2