Default Branch

33ca0dcb9d · ggml-hip : add -fno-finite-math-only alongside -ffast-math (#25373) · Updated 2026-07-07 11:27:50 +00:00

Branches

f249c997a8 · llama : adapt to F16 KQ_pos · Updated 2024-02-19 11:31:02 +00:00    kanshan

7699
62

412735ec70 · Merge branch 'master' into gg/metal-batched · Updated 2024-02-19 09:25:24 +00:00    kanshan

7712
6

47c662b0de · fix some spaces added by IDE in math op · Updated 2024-02-18 20:40:35 +00:00    kanshan

7722
4

974e3cadff · ggml : try another fix · Updated 2024-02-17 16:14:35 +00:00    kanshan

7741
2

e856bfed3b · hf : add support for --repo and --file · Updated 2024-02-15 13:05:15 +00:00    kanshan

7755
3

ccd757a174 · convert : fix mistakes from refactoring · Updated 2024-02-13 17:01:30 +00:00    kanshan

7763
4

5c977221d2 · iq1_s: slightly faster dot product · Updated 2024-02-13 13:18:27 +00:00    kanshan

7769
15

4246b71ad7 · Fix compiler warnings (shadow variable) · Updated 2024-02-13 06:44:56 +00:00    kanshan

7772
1

7286b83d3f · BERT WIP · Updated 2024-02-06 22:10:11 +00:00    kanshan

7825
1

adcf16fd68 · py : fix empty bytes arg · Updated 2024-02-05 17:53:07 +00:00    kanshan

7835
2

91c453fb11 · One cannot possibly be defining static_assert in a C++ compilation · Updated 2024-02-05 11:22:14 +00:00    kanshan

7840
2

49a483e0f2 · wip · Updated 2024-02-04 10:34:36 +00:00    kanshan

7866
60

a647257b47 · cuda : express strides with helper constants · Updated 2024-02-04 09:45:26 +00:00    kanshan

7866
60

b957b8f5f6 · cuda : add flash_attn kernel (wip) · Updated 2024-02-01 17:49:57 +00:00    kanshan

7870
39

ac26f27028 · cuda : increase C to 128 for better performance · Updated 2024-02-01 15:08:29 +00:00    kanshan

7870
61

1ad42b1f1e · ggml : ggml_soft_max uses F16 mask · Updated 2024-01-31 18:33:59 +00:00    kanshan

7870
36

719a087138 · iq3_xxs: forgotten update of the grid points · Updated 2024-01-30 16:39:07 +00:00    kanshan

7884
1

2bf91c5306 · metal : clean up · Updated 2024-01-25 11:29:45 +00:00    kanshan

7980
23

6ccbd1777a · wip · Updated 2024-01-24 13:45:04 +00:00    kanshan

7980
18

da23b56f25 · wip : no ic 8 step · Updated 2024-01-24 11:25:34 +00:00    kanshan

7980
18