Default Branch

c7fc509417 · opencl: add F32-to-I32 ARGMAX (#30897) · Updated 2026-10-22 09:02:50 +08:00

Branches

1ef3cc1a87 · imatrix : use GGUF regardless of the output filename · Updated 2025-07-24 11:22:41 +08:00    tqcq

5226
2

55cf48de1e · cuda : fix multi-seq, quantized FA · Updated 2025-07-23 01:48:53 +08:00    tqcq

5268
2

386892ec61 · sync : ggml · Updated 2025-07-19 16:46:12 +08:00    tqcq

5263
1

cfe5e98423 · graph : fix graph reuse reset of params · Updated 2025-07-18 22:50:32 +08:00    tqcq

5266
1

9106d7595d · model : fix build after merge conflict · Updated 2025-07-18 16:50:59 +08:00    tqcq

5269
1

05baa62a73 · kv-cache : fix k-shift for multiple streams · Updated 2025-07-18 01:18:36 +08:00    tqcq

5278
1

07908a824a · server : pre-calculate EOG logit biases · Updated 2025-07-16 18:47:05 +08:00    tqcq

5291
1

9f8d285901 · server : fix handling of the ignore_eos flag · Updated 2025-07-16 12:37:18 +08:00    tqcq

5296
1

f68669d50f · fix and opt kernel launch · Updated 2025-07-15 19:28:26 +08:00    tqcq

5331
3

942c55cd57 · imatrix : avoid using imatrix.dat in README · Updated 2025-07-13 04:50:10 +08:00    tqcq

5321
32

1180752835 · cuda : support Falcon-H1 state size for SSM_SCAN · Updated 2025-07-10 00:18:37 +08:00    tqcq

5351
1

4d6a179c68 · gguf-py : avoid adding duplicate tensor mappings for Jamba · Updated 2025-07-09 23:58:35 +08:00    tqcq

5351
61

b7c6ece5b5 · ggml-ci · Updated 2025-07-09 20:13:34 +08:00    tqcq

5356
24

7634d14d7a · test-model-random : fix seq_id buffer overflow · Updated 2025-07-09 06:23:58 +08:00    tqcq

5356
18

2ff3354c33 · memory : fix broken batch splits for recurrent cache · Updated 2025-07-08 09:23:14 +08:00    tqcq

5367
1

996195299e · up. · Updated 2025-07-08 05:42:40 +08:00    tqcq

5519
6

bf8b39015f · metal : reuse graphs · Updated 2025-07-08 02:37:07 +08:00    tqcq

5375
3

886da0a2c5 · kv-cache : prepare K/V buffers for separation · Updated 2025-07-04 15:13:16 +08:00    tqcq

5386
1

dfceb012ee · llama : add "virtual sequences" · Updated 2025-07-03 01:26:55 +08:00    tqcq

5397
8

71bef66591 · cuda : graceful fallback for Mamba-1 models with weird embd size · Updated 2025-07-02 15:49:36 +08:00    tqcq

5406
44