Default Branch

main b2cb4512c5 · Create parameters overview (#1269) · Updated 2026-02-20 07:20:56 +01:00

ZIP  TAR.GZ

Branches

ik/fix_again_cmake ffabdce3ce · Fighting with cmake · Updated 2025-03-22 16:49:43 +01:00    git	4211 3603	Compare	ZIP  TAR.GZ
ik/better_flash_mla 0964a49990 · Cleanup · Updated 2025-03-22 11:00:26 +01:00    git	4211 3605	Compare	ZIP  TAR.GZ
ik/gemma3 e1684a8d47 · Revert changes to convert_hf_to_gguf.py · Updated 2025-03-21 18:01:33 +01:00    git	4211 3604	Compare	ZIP  TAR.GZ
ik/bug_missing_parentheses 5e1944bdec · Fix bug: missing parentheses in logical expression · Updated 2025-03-21 13:17:48 +01:00    git	4211 3601	Compare	ZIP  TAR.GZ
ik/offline_repack_patterns da7d0ffba6 · Specify tensor name regex for tensors to be repacked · Updated 2025-03-21 08:03:13 +01:00    git	4211 3600	Compare	ZIP  TAR.GZ
ik/FlashMLA-3 4632cb94d8 · FlashMLA-3: the best of both worlds - CPU only · Updated 2025-03-20 15:58:19 +01:00    git	4211 3605	Compare	ZIP  TAR.GZ
ik/offline_repack 9fe6fc3782 · Add missing include · Updated 2025-03-20 15:57:38 +01:00    git	4211 3604	Compare	ZIP  TAR.GZ
ik/tensor_override_honor_mmap 1b62d0fae3 · Honor mmap setting when using tensor overrides · Updated 2025-03-19 16:05:04 +01:00    git	4211 3597	Compare	ZIP  TAR.GZ
ik/fix_dup_q 60c9495c2f · Fix ggml_compute_forward_dup_q · Updated 2025-03-19 15:44:34 +01:00    git	4211 3596	Compare	ZIP  TAR.GZ
ik/avoid_cuda_mla_1 529f75c220 · Prevent FlashMLA-1 from running on CUDA · Updated 2025-03-19 11:07:51 +01:00    git	4211 3595	Compare	ZIP  TAR.GZ
ik/mla2_q80_cache_cpu 96d1235fb0 · Allow q8_0 cache on the CPU for FlashMLA-2 · Updated 2025-03-18 13:08:52 +01:00    git	4211 3593	Compare	ZIP  TAR.GZ
ik/mla2_q80_cache a9440bd3e9 · Make Q8_0 KV cache work with mla=2,fa on CUDA · Updated 2025-03-18 10:57:32 +01:00    git	4211 3593	Compare	ZIP  TAR.GZ
ik/fix_pr_261 55b2cf98d2 · Fix #261 · Updated 2025-03-18 07:43:45 +01:00    git	4211 3592	Compare	ZIP  TAR.GZ
ik/use_bf16_when_no_mmq f326a5eaf7 · Compile time option to use bf16 for qunts without MMQ kernels · Updated 2025-03-17 19:38:20 +01:00    git	4211 3590	Compare	ZIP  TAR.GZ
ik/flash_mla2_cuda_no_f32 b147e31f5a · Reduce memory usage for FlashMLA-2 · Updated 2025-03-17 14:00:26 +01:00    git	4211 3601	Compare	ZIP  TAR.GZ
ik/prepare_wk_b f2fb15de77 · Fix CUDA · Updated 2025-03-16 06:40:18 +01:00    git	4211 3596	Compare	ZIP  TAR.GZ
ik/flash_mla2_no_f32 765c03d09b · FlashMLA-2: slightly smaller computer buffer size · Updated 2025-03-12 14:06:31 +01:00    git	4211 3590	Compare	ZIP  TAR.GZ
ik/cuda_flash_mla_q8_0 50bbc3f335 · FlashMLA(CUDA) - allow q8_0 for KV cache · Updated 2025-03-11 17:41:39 +01:00    git	4211 3590	Compare	ZIP  TAR.GZ
ik/flash_precision e0eebfd8ad · Try using fp32 for FlashMLA · Updated 2025-03-10 18:07:53 +01:00    git	4211 3587	Compare	ZIP  TAR.GZ
ik/mla_imatrix 56921ccd49 · imatrix: wv_b <-> wkv_b · Updated 2025-03-10 14:31:22 +01:00    git	4211 3589	Compare	ZIP  TAR.GZ