Commit Graph

107 Commits (10bbbed218458b8a899aac2140ec738d8d716f05)

Author SHA1 Message Date
Disty0 b601f0d402 SDNQ expose svd_steps and update module skip keys 2025-10-14 00:15:09 +03:00
Vladimir Mandic 2e4e741d47 seedvt2
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-10-12 15:35:08 -04:00
Disty0 df03ea9ba8 SDNQ add sdnq_post_load_quant and update Qwen keys 2025-10-08 00:29:36 +03:00
Vladimir Mandic 0092a8b86b add quantization_config for post-load
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-10-07 10:14:31 -04:00
Disty0 9e52d0c1fb SDNQ add SVDQuant quantization method 2025-10-05 22:50:30 +03:00
Vladimir Mandic 70a2c209b1 cleanup
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-09-28 20:22:25 -04:00
Disty0 a8b850adf4 move hf quantizer hijacks to sdnq 2025-09-12 20:54:44 +03:00
Vladimir Mandic 8ed04fb9a6 cleanup
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-09-11 13:02:52 -04:00
Vladimir Mandic 7940217764 add models_not_to_quant option
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-09-11 12:31:07 -04:00
Disty0 bc7c89c070 add typing to sdpa hijacks 2025-09-10 04:43:57 +03:00
Disty0 251a1ce3d9 fix modules_to_not_convert and modules_dtype_dict not resetting 2025-09-10 04:20:57 +03:00
Disty0 dcaeed360d SDNQ set return device to gpu with shuffle weights option 2025-09-08 18:56:29 +03:00
Vladimir Mandic 78c2a629b6 add experimental tensorrt quantization
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-09-05 10:43:05 -04:00
Vladimir Mandic 2124ab6879 trt experiment
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-09-04 10:53:32 -04:00
Disty0 e49814098e Add sdnq_modules_dtype_dict 2025-08-20 14:58:54 +03:00
Disty0 0946710662 Add sdnq_modules_to_not_convert to UI settings 2025-08-20 04:38:20 +03:00
Disty0 8ca74d0cd2 SDNQ rename unused param_name arg to op 2025-08-13 22:10:30 +03:00
Disty0 15cb8fe9f8 SDNQ add modules_dtype_dict and fix Qwen Image with quants less than 5 bits 2025-08-13 00:07:36 +03:00
Vladimir Mandic 87bd347116 cleanup flux loader
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-08-11 16:05:39 -04:00
Disty0 afb3a5a06d SDNQ move non_blocking to quant config 2025-08-11 15:07:02 +03:00
Vladimir Mandic 6a6605191f configurable image fit in all image views
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-08-06 11:33:50 -04:00
Vladimir Mandic 7bba30e797 sdnq obey diffusers_to_gpu
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-08-06 10:12:27 -04:00
Vladimir Mandic ba4bff08d6 remove ldsr and refactor sdnq device map
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-08-06 09:27:54 -04:00
Vladimir Mandic 4be093b80f add diffusers_offload_nonblocking setting
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-08-01 16:38:31 -04:00
Vladimir Mandic b291c337a1 refactor internal post loop
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-08-01 10:45:39 -04:00
Vladimir Mandic fa44521ea3 offload-never and offload-always per-module and new highvram profile
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-07-31 11:40:24 -04:00
Vladimir Mandic 052f097956 lint
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-07-30 13:14:04 -04:00
Disty0 f02cfeaef9 Rename SDNQ TE dtype default to Same as model 2025-07-27 23:06:41 +03:00
Vladimir Mandic b5a87c4828 modify installer checks
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-07-26 18:40:28 -04:00
Vladimir Mandic ed1e59464e update requirements
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-07-06 14:53:25 -04:00
Vladimir Mandic 2b9056179d add lbm background replace with relightining
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-07-04 15:33:16 -04:00
Disty0 4aeb9f004c Cleanup layerwise 2025-07-04 06:12:25 +03:00
Disty0 9d5571f4f8 Use layerwise casting with Flux FP8 models 2025-07-04 04:26:43 +03:00
Disty0 fe9a3b8506 Add device map support to Flux and Chroma and add custom UNet support to Chroma 2025-07-04 02:22:42 +03:00
Disty0 cf90e5621a Add _skip_layerwise_casting_patterns to SDNQ skip list 2025-07-04 00:04:01 +03:00
Vladimir Mandic c4d9338d2e major refactoring of modules
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-07-03 09:18:38 -04:00
Disty0 71f7474de2 Unify quant options 2025-06-27 21:05:14 +03:00
Disty0 3b8ced444c Add auto quantization mode 2025-06-27 18:54:15 +03:00
Disty0 0e4d712f27 Chroma fixes 2025-06-26 21:25:50 +03:00
Enes Sadık Özbek e91208bea9
Merge branch 'dev' into feature/chroma-support 2025-06-26 17:02:00 +03:00
Disty0 0f6eb624c9 Use llm_int8_skip_modules with bnb 2025-06-26 03:10:26 +03:00
Disty0 dc8fd006b2 Add modules_to_not_convert to pre-mode quants 2025-06-26 02:47:10 +03:00
Enes Sadık Özbek 21bdde12d3
Merge branch 'dev' into feature/chroma-support 2025-06-26 01:56:34 +03:00
Disty0 c25d15398f SDNQ pre-mode don't quant if weights_dtype == 'none' 2025-06-26 01:48:01 +03:00
Disty0 81e55f0459 Fix SDNQ pre-mode 2025-06-25 22:58:40 +03:00
Vladimir Mandic 5b486a6ef1 sdnq add xyz grid support, improve offloading compatibility
Signed-off-by: Vladimir Mandic <mandic00@live.com>
2025-06-25 15:32:37 -04:00
Disty0 87e6d3f4fc SDNQ add modules_to_not_convert and don't quant _keep_in_fp32_modules layers in post mode 2025-06-20 20:41:11 +03:00
Disty0 2c4850cc2b Log more info with SDNQ 2025-06-18 18:03:46 +03:00
Disty0 86cd272b96 SDNQ fix Dora 2025-06-18 16:24:42 +03:00
Enes Sadık Özbek 4b3ce06916
Initial support for Chroma 2025-06-18 00:38:17 +00:00