Default Branch

main f96eaddba8 · Revert DFlash SWA optimization (#2039) · Updated 2026-06-26 04:00:09 -05:00

ZIP  TAR.GZ

Branches

ik/bitnet_optional_scales a3fe796f6c · Bitnet: make the scale tensors optional · Updated 2024-10-19 11:37:33 -05:00    jdelony	4666 3467		ZIP  TAR.GZ
ik/cached_graph 0e76d21b96 · Adding agray3's graph caching approach · Updated 2024-10-18 10:01:08 -05:00    jdelony	4666 3465		ZIP  TAR.GZ
ik/fix_reduce_windows e732da1f57 · Attempt to blindly fix Windows build failure · Updated 2024-10-18 04:35:47 -05:00    jdelony	4666 3465		ZIP  TAR.GZ
ik/iq4_knn c4292bf2d9 · iq4_knn: Metal - predictably bad · Updated 2024-10-18 03:48:00 -05:00    jdelony	4666 3468		ZIP  TAR.GZ
ik/iq4_kss 9612cd79d6 · iq4_kss: very slightly faster Metal dot product · Updated 2024-10-16 07:08:15 -05:00    jdelony	4666 3473		ZIP  TAR.GZ
ik/metal_faster_iq4ks 3e0c2519d3 · iq4_ks: faster dot product on Metal · Updated 2024-10-16 06:04:59 -05:00    jdelony	4666 3462		ZIP  TAR.GZ
ik/metal_fix_iq3k 55f91a98f1 · iq3_k: slightly faster Metal dot product · Updated 2024-10-14 02:41:26 -05:00    jdelony	4666 3461		ZIP  TAR.GZ
ik/metal_fix_iq2k f74905d649 · iq2_k: optimize Metal dot product · Updated 2024-10-13 06:09:53 -05:00    jdelony	4666 3461		ZIP  TAR.GZ
ik/iq2k_experiments f9f15c27b6 · iq2_ks: faster Metal · Updated 2024-10-13 04:23:14 -05:00    jdelony	4666 3470		ZIP  TAR.GZ
ik/better_model_info e441c897a4 · Better model info · Updated 2024-10-10 09:38:59 -05:00    jdelony	4666 3456		ZIP  TAR.GZ
ik/iq3_ks e734e888e1 · iq3_ks: AVX2 · Updated 2024-10-10 02:48:42 -05:00    jdelony	4666 3463		ZIP  TAR.GZ
ik/iq4_k_xxs f61c37967a · iq3_kl: use iq4_ks instead of iq4_k/iq4_xs · Updated 2024-10-09 04:50:43 -05:00    jdelony	4666 3467		ZIP  TAR.GZ
ik/qstats df2bd86a31 · WIP · Updated 2024-10-06 01:09:51 -05:00    jdelony	4666 3458		ZIP  TAR.GZ
ik/cleanup_fudge_factors acaa4869af · Move scale fudge factors to quantization · Updated 2024-10-04 08:14:52 -05:00    jdelony	4666 3453		ZIP  TAR.GZ
ik/cpp_17 a553eb191a · Make the entire project c++17 · Updated 2024-10-04 06:23:21 -05:00    jdelony	4666 3453		ZIP  TAR.GZ
ik/skip_unnecessary_quantize ed477f1cdc · Do not quantize activations if not necessary also for MoE models · Updated 2024-10-04 03:11:02 -05:00    jdelony	4666 3452		ZIP  TAR.GZ
ik/faster_q60_avx2 38eb7fa499 · q6_0: this is slightly better · Updated 2024-10-02 10:07:55 -05:00    jdelony	4666 3451		ZIP  TAR.GZ
ik/fused_mul_unary a8e932b734 · Fused y*unary(x) op: Metal · Updated 2024-10-02 08:51:29 -05:00    jdelony	4666 3452		ZIP  TAR.GZ
ik/add_q60 037bbd2d58 · q6_0: can now be used for kv-cache on Metal · Updated 2024-10-02 06:54:25 -05:00    jdelony	4666 3457		ZIP  TAR.GZ
ik/faster_iq4nl_quantize 1fb3115412 · iq4_nl: faster quantization · Updated 2024-10-01 23:43:09 -05:00    jdelony	4666 3447		ZIP  TAR.GZ