Commit Graph

  • 81266c22d6 iq6_k: WIP (nothing works) Kawrakow 2024-08-01 16:08:32 +03:00
  • cfb0410067 iq6_k: WIP (nothing works) Iwan Kawrakow 2024-08-01 16:08:32 +03:00
  • bf745357ec Fix Zen4 implementation of iq3_k, iq4_k, iq5_k ik/iq6_k Iwan Kawrakow 2024-08-09 10:32:07 +03:00
  • 3ef9f7a509 iq6_k: AVX2 Iwan Kawrakow 2024-08-09 06:58:55 +03:00
  • c9c5ed5793 iq6_k: Metal Iwan Kawrakow 2024-08-08 16:27:43 +02:00
  • b8abe10b6d iq6_k: NEON Iwan Kawrakow 2024-08-08 15:51:10 +02:00
  • 0d55d5f078 iq6_k: slightly better Zen4 iqk_mul_mat Iwan Kawrakow 2024-08-08 14:01:08 +03:00
  • f3a823ce72 iq6_k: Zen4 iqk_mul_mat Iwan Kawrakow 2024-08-08 11:17:42 +03:00
  • 07d3b4caec iq6_k: CUDA dot product Iwan Kawrakow 2024-08-07 19:24:09 +03:00
  • b3d6e10c7d iq6_k: CUDA dequantize Iwan Kawrakow 2024-08-07 17:25:21 +03:00
  • 85f448e2b1 iq6_k: WIP (quantize/dequantize) Iwan Kawrakow 2024-08-07 16:49:43 +03:00
  • 54ce23bb61 iq6_k: WIP (nothing works) Iwan Kawrakow 2024-08-01 16:08:32 +03:00
  • 58a323f585 Adding IQ2_TN for use with ternary models (#13) Kawrakow 2024-08-07 07:56:09 +02:00
  • a9f302ebe2 Adding IQ2_TN for use with ternary models (#13) Kawrakow 2024-08-07 07:56:09 +02:00
  • 8178075f84 iq2_tn: small NEON improvement ik/iq2_tn Iwan Kawrakow 2024-08-06 12:08:22 +02:00
  • 9780ac4591 iq2_tn: AVX2 PP improvement Iwan Kawrakow 2024-08-06 12:34:44 +03:00
  • 2cc6338670 iq2_tn: CUDA Iwan Kawrakow 2024-08-06 10:06:39 +03:00
  • 5d02f7f4a5 iq2_tn: Metal Iwan Kawrakow 2024-08-06 07:39:39 +02:00
  • e528505fc8 iq2_tn: NEON Iwan Kawrakow 2024-08-06 06:19:59 +02:00
  • 810285581c iq2_tn: initial NEON version Iwan Kawrakow 2024-08-05 17:53:02 +02:00
  • a63ba11a25 iq2_tn: AVX2 Iwan Kawrakow 2024-08-05 17:25:14 +03:00
  • d0cc103878 iq2_tn: AVX512 Iwan Kawrakow 2024-08-05 15:47:12 +03:00
  • c063954c1a iq2_tn: AVX512 Iwan Kawrakow 2024-08-05 15:17:27 +03:00
  • dd0b08d1d8 iq2_tn: AVX512 Iwan Kawrakow 2024-08-05 14:53:49 +03:00
  • 1b41d792ec iq2_tn: TriLM specific 2.0625 bpw quantization Iwan Kawrakow 2024-08-05 14:22:05 +03:00
  • 695c7eef49 q2_K: allow it to detect ternary nets and quantize accordingly Kawrakow 2024-08-05 11:59:36 +03:00
  • b409c15363 q2_K: allow it to detect ternary nets and quantize accordingly Iwan Kawrakow 2024-08-05 11:59:36 +03:00
  • 54481695a0 q2_K: allow it to detect ternary nets and quantize accordingly ik/trinet Iwan Kawrakow 2024-08-05 11:59:36 +03:00
  • 74f2f50abf Update README.md Kawrakow 2024-08-05 07:35:30 +02:00
  • c11c7c8cae Update README.md Kawrakow 2024-08-05 07:35:30 +02:00
  • 8db9c5b1fd iq3_k, iq5_k: faster quantization Kawrakow 2024-08-05 08:13:53 +03:00
  • 6901b3bf14 iq3_k, iq5_k: faster quantization Iwan Kawrakow 2024-08-05 08:13:53 +03:00
  • de818b77d6 iq3_k, iq5_k: faster quantization ik/faster_iq3_iq5_quantize Iwan Kawrakow 2024-08-05 08:13:53 +03:00
  • 5b06d7999c iq4_k: speedup quantization by a factor of ~2 Kawrakow 2024-08-03 19:32:43 +03:00
  • e830f4a5f7 iq4_k: speedup quantization by a factor of ~2 Iwan Kawrakow 2024-08-03 19:32:43 +03:00
  • 30c002d22d iq4_k: speedup quantization by a factor of ~2 ik/faster_iq4k_quantize Iwan Kawrakow 2024-08-03 19:32:43 +03:00
  • 2af0d6fbac Add copyright notice Kawrakow 2024-07-31 15:06:30 +03:00
  • 3d1446b937 Add copyright notice Iwan Kawrakow 2024-07-31 15:06:30 +03:00
  • 904fdbcfb7 iq2/3_k: tiny bit faster Metal dot products Kawrakow 2024-07-31 14:53:54 +02:00
  • b572dd5347 iq2/3_k: tiny bit faster Metal dot products Iwan Kawrakow 2024-07-31 14:53:54 +02:00
  • 088a8360a1 iq3_k: slightly faster Metal dequantize kernel Kawrakow 2024-07-31 08:49:44 +02:00
  • 394ed3913c iq3_k: slightly faster Metal dequantize kernel Iwan Kawrakow 2024-07-31 08:49:44 +02:00
  • 606f02ae89 iq3_k: Metal dot product Kawrakow 2024-07-31 08:44:19 +02:00
  • 062313dab4 iq3_k: Metal dot product Iwan Kawrakow 2024-07-31 08:44:19 +02:00
  • 95a6820d79 iq2_k: Metal dot product finally works Kawrakow 2024-07-31 08:01:45 +02:00
  • 57df5ccdd7 iq2_k: Metal dot product finally works Iwan Kawrakow 2024-07-31 08:01:45 +02:00
  • 033299c9f9 iq3_k: Metal dequantize Kawrakow 2024-07-30 20:17:58 +02:00
  • 30d2d1b1eb iq3_k: Metal dequantize Iwan Kawrakow 2024-07-30 20:17:58 +02:00
  • 2927d4f841 iq3_k: NEON Kawrakow 2024-07-30 19:17:12 +02:00
  • 4c2c644dcc iq3_k: NEON Iwan Kawrakow 2024-07-30 19:17:12 +02:00
  • 9c1eea6048 iq3_k: AVX2 iqk_mul_mat Kawrakow 2024-07-30 19:01:35 +03:00
  • 93d09d1935 iq3_k: AVX2 iqk_mul_mat Iwan Kawrakow 2024-07-30 19:01:35 +03:00
  • a9fa3b1563 iq3_k: AVX512 iqk_mul_mat Kawrakow 2024-07-30 18:40:10 +03:00
  • 9d0cf7a399 iq3_k: AVX512 iqk_mul_mat Iwan Kawrakow 2024-07-30 18:40:10 +03:00
  • a4371b7842 iq3_k: faster CUDA dot product Kawrakow 2024-07-30 17:18:31 +03:00
  • fd1ae85a32 iq3_k: faster CUDA dot product Iwan Kawrakow 2024-07-30 17:18:31 +03:00
  • 81f15c0ba8 iq3_k: CUDA dot product Kawrakow 2024-07-30 16:57:39 +03:00
  • 0d19d19af8 iq3_k: CUDA dot product Iwan Kawrakow 2024-07-30 16:57:39 +03:00
  • fb4cff3458 iq3_k: Basics Kawrakow 2024-07-30 16:11:25 +03:00
  • 4f237d44f6 iq3_k: Basics Iwan Kawrakow 2024-07-30 16:11:25 +03:00
  • 7dcd64c9bd iq2_k: very slightly better CUDA dot product Kawrakow 2024-07-30 15:17:21 +03:00
  • 36204c4ec7 iq2_k: very slightly better CUDA dot product Iwan Kawrakow 2024-07-30 15:17:21 +03:00
  • 0c1d7383a5 iq2_k: better CUDA dot product Kawrakow 2024-07-30 12:49:55 +03:00
  • e950b17125 iq2_k: better CUDA dot product Iwan Kawrakow 2024-07-30 12:49:55 +03:00
  • f30bcc1e17 iq2_k: CUDA dot product finally works Kawrakow 2024-07-30 12:33:48 +03:00
  • ab4f9e1fdb iq2_k: CUDA dot product finally works Iwan Kawrakow 2024-07-30 12:33:48 +03:00
  • 53fdb30ca6 iq5_k: CUDA dot product finally works Kawrakow 2024-07-30 11:22:08 +03:00
  • 69842c6ad8 iq5_k: CUDA dot product finally works Iwan Kawrakow 2024-07-30 11:22:08 +03:00
  • 8654a425ae Factor out iqk CUDA dot products Kawrakow 2024-07-30 09:14:18 +03:00
  • f6813cac0e Factor out iqk CUDA dot products Iwan Kawrakow 2024-07-30 09:14:18 +03:00
  • 99456e2e94 iq5_k: CUDA dot product still not working Kawrakow 2024-07-29 21:01:05 +03:00
  • 22d1568c1c iq5_k: CUDA dot product still not working Iwan Kawrakow 2024-07-29 21:01:05 +03:00
  • b591023479 iq5_k: Metal Kawrakow 2024-07-29 16:47:06 +02:00
  • d8d022a01b iq5_k: Metal Iwan Kawrakow 2024-07-29 16:47:06 +02:00
  • 0ab3f0ff86 iq5_k: NEON Kawrakow 2024-07-29 13:57:14 +02:00
  • bd36ade98d iq5_k: NEON Iwan Kawrakow 2024-07-29 13:57:14 +02:00
  • daf608e227 iq5_k: AVX512 Kawrakow 2024-07-29 14:26:14 +03:00
  • c0d0607f19 iq5_k: AVX512 Iwan Kawrakow 2024-07-29 14:26:14 +03:00
  • e9c3ebcbe9 iq5_k: AVX2 Kawrakow 2024-07-29 13:39:50 +03:00
  • c56ddee38c iq5_k: AVX2 Iwan Kawrakow 2024-07-29 13:39:50 +03:00
  • e5cd93b4b7 iq5_k: Basics Kawrakow 2024-07-29 12:38:46 +03:00
  • 5d341757bc iq5_k: Basics Iwan Kawrakow 2024-07-29 12:38:46 +03:00
  • ace8f921bb iq2_k: Metal. Dot product is wrong Kawrakow 2024-07-29 09:49:32 +02:00
  • 06e255ac9d iq2_k: Metal. Dot product is wrong Iwan Kawrakow 2024-07-29 09:49:32 +02:00
  • f7ab9a13df iq2_k: NEON Kawrakow 2024-07-29 07:26:36 +02:00
  • f476ea3b50 iq2_k: NEON Iwan Kawrakow 2024-07-29 07:26:36 +02:00
  • cc8e351b68 iq2_k: slightly faster AVX512 Kawrakow 2024-07-29 06:51:44 +03:00
  • c0fe03b5c8 iq2_k: slightly faster AVX512 Iwan Kawrakow 2024-07-29 06:51:44 +03:00
  • 764d4675b8 iq2_k: simplify AVX512 Kawrakow 2024-07-28 21:05:56 +03:00
  • 7d08719975 iq2_k: simplify AVX512 Iwan Kawrakow 2024-07-28 21:05:56 +03:00
  • 21319d6fca iq2_k: AVX2 Kawrakow 2024-07-28 20:50:21 +03:00
  • 13091d39e8 iq2_k: AVX2 Iwan Kawrakow 2024-07-28 20:50:21 +03:00
  • 3f7dad3000 iq2_k: Basics Kawrakow 2024-07-28 19:43:18 +03:00
  • c85e139c68 iq2_k: Basics Iwan Kawrakow 2024-07-28 19:43:18 +03:00
  • 7b3b413fe0 Add copyright notice ik/iq2_k Iwan Kawrakow 2024-07-31 15:06:30 +03:00
  • aeded5ec87 iq2/3_k: tiny bit faster Metal dot products Iwan Kawrakow 2024-07-31 14:53:54 +02:00
  • 09c94ae948 iq3_k: slightly faster Metal dequantize kernel Iwan Kawrakow 2024-07-31 08:49:44 +02:00
  • 0b9177bb11 iq3_k: Metal dot product Iwan Kawrakow 2024-07-31 08:44:19 +02:00
  • eeaa65d5b7 iq2_k: Metal dot product finally works Iwan Kawrakow 2024-07-31 08:01:45 +02:00
  • 3737acf6b6 iq3_k: Metal dequantize Iwan Kawrakow 2024-07-30 20:17:58 +02:00