Default Branch

3a945af45d · Faster prompt processing on CUDA (#1687) · Updated 2026-04-25 07:05:23 +00:00

Branches

2565b29f33 · Handle incompatible DeepSeek GGUFs · Updated 2025-05-07 15:25:40 +00:00    ikawrakow

4437
3673

93d053f7ab · Fix DeepSeek q8_0 cache · Updated 2025-05-07 09:02:05 +00:00    ikawrakow

4437
3670

e6da985f02 · Fix build for Xeon Gold 6226R · Updated 2025-05-07 07:23:18 +00:00    ikawrakow

4437
3669

1982beb005 · Minor tweak · Updated 2025-05-07 06:07:34 +00:00    ikawrakow

4437
3674

296367a50d · Update vocab.py · Updated 2025-05-05 06:37:01 +00:00    ikawrakow

4437
3672

f455ead8aa · Fix DeepSeek FA · Updated 2025-05-05 05:31:55 +00:00    ikawrakow

4437
3667

a3975acd4c · Add batch warmup to sweep-bench · Updated 2025-05-04 08:21:19 +00:00    ikawrakow

4437
3664

3498ea4228 · CUDA: MMQ for iq4_ks now works · Updated 2025-05-04 06:19:23 +00:00    ikawrakow

4437
3666

5782f1bdf0 · Yet another · Updated 2025-05-03 17:00:20 +00:00    ikawrakow

4437
3663

056f08182a · Use MMA for TG also when quantized · Updated 2025-05-03 12:34:56 +00:00    ikawrakow

4437
3661

267a12aaa0 · Trying to fix iq1_s_r4/iq1_m_r4 quantization failure · Updated 2025-05-03 10:53:39 +00:00    ikawrakow

4437
3660

0e247afcac · Fix model architecture name · Updated 2025-05-02 02:43:18 +00:00    ikawrakow

4437
3658

2b7061967a · Also this was wrong · Updated 2025-05-01 16:05:08 +00:00    ikawrakow

4437
3659

a0d10704cd · Dynamic Yarn · Updated 2025-05-01 12:29:09 +00:00    ikawrakow

4437
3658

6c70182744 · Updates · Updated 2025-04-30 13:10:45 +00:00    ikawrakow

4437
3654

b05c85e487 · Make it also work, not just compile · Updated 2025-04-30 08:45:07 +00:00    ikawrakow

4437
3657

b036119637 · Add missing enum values for qwen3 and qwen3moe · Updated 2025-04-29 08:04:38 +00:00    ikawrakow

4437
3655

1f77976476 · Update README.md · Updated 2025-04-28 14:25:48 +00:00    ikawrakow

4437
3652

20d50172d0 · Much better FA TG with q8_0 KV cache · Updated 2025-04-28 08:26:28 +00:00    ikawrakow

4437
3667

957308ca09 · Fix division by zero bug · Updated 2025-04-26 07:08:37 +00:00    ikawrakow

4437
3650