Default Branch

9d07d8681e · perplexity: fix int overflows in large context x vocab buffer sizing (#2150) · Updated 2026-07-18 06:13:47 +00:00

Branches

fb26910109 · Take into account swiglu limits · Updated 2026-07-18 15:49:42 +00:00

7
38

f535d8d421 · Fix race in indexer topk on CUDA · Updated 2026-07-17 15:05:15 +00:00

4
1

1e5795167f · Add forgotten check · Updated 2026-07-15 10:51:39 +00:00

10
2

d477e24921 · Minor · Updated 2026-07-14 08:42:54 +00:00

14
4

49333012ae · fix --skip-chat-parsing and --no-prefill-assistant consume the next argument · Updated 2026-07-14 01:59:06 +00:00

14
1

618930b4b0 · Merge remote-tracking branch 'origin/main' into ik/better_indexer_mask · Updated 2026-07-12 16:37:38 +00:00

18
11

b6cdcced6f · Merge remote-tracking branch 'origin/main' into ik/glm_dsa_cuda · Updated 2026-07-12 05:10:28 +00:00

22
9

ff27465bac · Fix bug. Fix suggested by @sayap using GLM-5.2 · Updated 2026-07-11 06:52:20 +00:00

33
7

bdb927cf4b · Make it a command line option · Updated 2026-07-08 06:58:50 +00:00

38
2

66d2e14e97 · Fix #2093 · Updated 2026-07-07 16:20:12 +00:00

40
1

be1f64b6e6 · GLM DSA: reduce indexer cache size · Updated 2026-07-06 17:36:26 +00:00

44
1

944ffa98f5 · Be able to force synchronization when copying graph ginputs · Updated 2026-07-06 17:02:43 +00:00

47
1

e1c954177d · GLM-DSA: do not compute indexer score if context < n_top_k · Updated 2026-07-06 12:35:46 +00:00

47
1

14118a4697 · Merge remote-tracking branch 'origin/main' into ik/glm_dsa_quantized_indexer_cache · Updated 2026-07-06 10:01:57 +00:00

49
5

c3488bc346 · Cleanup · Updated 2026-07-06 09:24:55 +00:00

50
2

8d56a985e4 · Fix the fix · Updated 2026-07-06 09:04:57 +00:00

54
3

d529cf86a7 · Fix Windows build after CUB addition · Updated 2026-07-03 15:52:49 +00:00

54
1

8c52b867e6 · GLM-DSA: improve TG performance · Updated 2026-07-02 09:40:31 +00:00

58
1

f11eb29190 · GLM-DSA: minor optimization · Updated 2026-07-02 08:24:04 +00:00

59
1

06e6530d32 · Various · Updated 2026-07-01 13:31:54 +00:00

61
19