Commit Graph

45 Commits (39ce3d7ea31b0ba51a861298a0a3ffa6607f8a2f)

Author SHA1 Message Date
Vladimir Mandic e5c494f999 cleanup logger 2026-02-19 11:09:13 +01:00
Vladimir Mandic a3074baf8b unified logger 2026-02-19 09:46:42 +01:00
Vladimir Mandic bfe014f5da modernize typing 2026-02-19 09:15:37 +01:00
awsr 09fdda05a4
Move to `modules` 2026-01-24 02:16:05 -08:00
awsr 82361e6633
Adjust names 2026-01-23 22:29:00 -08:00
awsr 58c3aecc00
Allow multiple identifiers for ErrorLimiter.notify
- Update identifiers.
- Also minor message formatting update.
2026-01-23 16:50:52 -08:00
awsr 3343d2e05f
Update and rewrite to use contextlib 2026-01-23 04:56:27 -08:00
awsr 65d8c9e7f2
Implement limiting system for excessive errors 2026-01-22 03:37:52 -08:00
Disty0 259a38a2ed fix sdnq lora 2025-12-27 23:07:53 +03:00
Disty0 b6e9332cfe SDNQ de-couple matmul dtype and add fp16 matmul 2025-11-22 02:16:20 +03:00
Disty0 3fbfae5963 cleanup 2025-11-18 02:37:10 +03:00
Disty0 524e92eee2 SDNQ fix Loras 2025-11-18 01:47:35 +03:00
Disty0 6f33ec3357 SDNQ use the model quant params instead of user settings on Lora 2025-11-10 00:12:38 +03:00
Vladimir Mandic ba270db6ad separate settings for lora fuse
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-11-08 11:08:06 -05:00
Disty0 b601f0d402 SDNQ expose svd_steps and update module skip keys 2025-10-14 00:15:09 +03:00
Disty0 5c042c5fb8 cleanup 2025-10-06 11:30:26 +03:00
Vladimir Mandic a315a004e9 linting
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-10-05 20:25:33 -04:00
Disty0 9e52d0c1fb SDNQ add SVDQuant quantization method 2025-10-05 22:50:30 +03:00
Disty0 54acf1760b Make SDNQ scales compatible with balanced offload 2025-10-03 18:13:55 +03:00
Disty0 afb3a5a06d SDNQ move non_blocking to quant config 2025-08-11 15:07:02 +03:00
Disty0 86cd272b96 SDNQ fix Dora 2025-06-18 16:24:42 +03:00
Disty0 25fc0094a9 SDNQ use quantize_device and return_device args and fix decompress_fp32 always being on 2025-06-14 21:29:08 +03:00
Disty0 2ba64abcde Cleanup 2025-06-14 00:54:18 +03:00
Disty0 5e013fb154 SDNQ optimize input quantization and use the word quantize instead of compress 2025-06-12 12:06:57 +03:00
Disty0 5eed9135e3 Split SDNQ into multiple files and linting 2025-06-10 03:18:25 +03:00
Disty0 976f0ba61f Cleanup 2025-06-05 20:59:58 +03:00
Disty0 90324f9c8c SDNQ fix lora with quant matmul 2025-05-29 18:25:12 +03:00
Disty0 dece497f10 Refactor SDNQ to use weights_dtype and rename decompress_int8_matmul to use_quantized_matmul 2025-05-27 15:49:21 +03:00
Disty0 280be31883 SDNQ fix Lora change 2025-05-27 00:08:32 +03:00
Disty0 84ddfb2868 SDNQ fix lora apply 2025-05-26 22:39:20 +03:00
Disty0 687c50dcc8 SDNQ fix Lora 2025-05-26 19:48:45 +03:00
Disty0 91bb07f650 SDNQ remove unused args and simplify decompressors 2025-05-26 15:51:53 +03:00
Disty0 4453efee76 Rename NNCF to SDNQ and rename quant schemes 2025-05-26 02:39:51 +03:00
Disty0 2d79380bd7 NNCF implement better layer hijacks and remove all NNCF imports 2025-05-26 01:12:28 +03:00
Vladimir Mandic 5c0e3b635c update diffusers and lint/changelog/todo
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-05-14 13:30:52 -04:00
Disty0 f4e3a81a84 NNCF experimental direct INT8 MatMul support 2025-05-12 21:41:49 +03:00
Disty0 9cfdc3c079 Remove NNCF device hijack 2025-05-11 18:30:10 +03:00
Disty0 75d169bc1c Fix NNCF Lora with model offload 2025-04-23 17:13:08 +03:00
Disty0 f1d8543cae NNCF lora support 2025-04-23 15:44:09 +03:00
Vladimir Mandic 84a24fb681 lora restore weights to orig device on apply
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-04-07 07:49:18 -04:00
Vladimir Mandic d30b1cb1c8 lora improvements
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-04-06 13:02:58 -04:00
Vladimir Mandic 8725cfc488 lora obey device
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-04-03 14:31:39 -04:00
Vladimir Mandic 5c6c1465f4 fix style apply params
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-04-03 10:03:48 -04:00
Vladimir Mandic 6430f7006f add monitor cli option and finish lora refactor
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-04-01 13:39:47 -04:00
Vladimir Mandic b5031a5eba lora modularize code
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-04-01 13:39:47 -04:00