Commit Graph

233 Commits (0bdbf300ac347640e5b81dff2fa187bcfe05574c)

Author SHA1 Message Date
Disty0 f35e644d44 cast svd to torch_dtype 2026-03-11 14:50:14 +03:00
Disty0 e202d43801 SDNQ set sdnq_dequantize_fp32 to True by default 2026-03-11 14:35:11 +03:00
Disty0 30d21f420f Allow all types in dyn quant 2026-03-11 13:46:24 +03:00
Disty0 b02a63c60c SDNQ update minimum dtype check 2026-03-11 04:12:37 +03:00
Disty0 76b184a9fe fix fp16 mm is selected instead of fp8 mm 2026-03-11 03:55:43 +03:00
Disty0 d9e628574a SDNQ add 15, 13, 11 and 9 bit support 2026-03-11 03:39:02 +03:00
Disty0 9fbe70195f SDNQ add 14, 12 and 10 bit support 2026-03-09 23:05:08 +03:00
Disty0 0160f6b3ef add sdnq variants of pytorch native fp formats too 2026-02-24 20:08:41 +03:00
Disty0 78efbc7e85 update sdnq 2026-02-24 19:47:30 +03:00
Vladimir Mandic bfe014f5da modernize typing 2026-02-19 09:15:37 +01:00
Disty0 f439d51ea7 SDNQ update CosmosTransformer3DModel keys 2026-02-04 10:19:23 +03:00
Disty0 df30ba28ac SDNQ clamp weight_std and add CosmosTransformer3DModel keys 2026-02-04 00:31:00 +03:00
Disty0 ced81d630d cleanup 2026-02-02 20:30:52 +03:00
Disty0 90cf83b90e update sdnq 2026-02-02 20:22:29 +03:00
vladmandic a4671045b6 lint and crlf
Signed-off-by: vladmandic <mandic00@live.com>
2026-01-24 10:28:46 +01:00
Disty0 8d6bfcd827 Update SDNQ 2026-01-23 14:39:07 +03:00
Disty0 784cda80aa update sdnq 2026-01-14 16:23:26 +03:00
Disty0 47dcab3522 update sdnq 2026-01-09 00:34:32 +03:00
Seunghoon Lee 49965dfda8
get_hip_arch_name -> get_hip_agent, use amdhip64_7.dll served within rocm package 2026-01-03 21:00:36 +09:00
vladmandic b9c18452f2 unify hip get arch name
Signed-off-by: vladmandic <mandic00@live.com>
2026-01-03 08:22:19 +01:00
vladmandic 4e8b0f83b4 lint
Signed-off-by: vladmandic <mandic00@live.com>
2026-01-01 16:33:49 +01:00
Disty0 8e34866238 SDNQ fix outdated PyTorch 2025-12-30 21:29:41 +03:00
Disty0 5e934a12a2 sdnq cleanup unused args 2025-12-28 20:08:58 +03:00
Disty0 b852ff42ef SDNQ fix wrong fp8 mm type is set 2025-12-27 17:27:05 +03:00
Disty0 db59d2b507 SDNQ handle packed floats in fp mm 2025-12-27 16:29:18 +03:00
Disty0 448e7b7735 replace the default fp6 type 2025-12-27 02:10:12 +03:00
Disty0 761fb82685 fix missing comma 2025-12-26 21:27:57 +03:00
Disty0 22b9e69a3e cleanup whitespace 2025-12-26 21:18:56 +03:00
Disty0 fd6c89a626 cleanup 2025-12-26 21:16:55 +03:00
Disty0 e7fa690321 cleanup 2025-12-26 20:10:55 +03:00
Disty0 4a4784eafa SDNQ add new stack of custom floating point types and remove irrelevant qtypes from the ui list 2025-12-26 20:09:17 +03:00
Disty0 471b6dc1b7 SDNQ add siglip_embedder to ZImage skip keys 2025-12-23 04:32:54 +03:00
Disty0 ce8b6d138c SDNQ remove forced uint4 from convs and cleanup 2025-12-13 01:32:52 +03:00
Disty0 de5d4f0165 SDNQ fix sr not doing anything 2025-12-09 19:57:34 +03:00
Disty0 949ff04577 SDNQ fix fp16 mm with fp8 weights and improve stochastic rounding performance 2025-12-09 17:41:29 +03:00
Disty0 1c2a81ee2d Make SDNQDequantizer a dataclass 2025-12-08 22:29:45 +03:00
vladmandic 69f0d6bf5d lint
Signed-off-by: vladmandic <mandic00@live.com>
2025-12-08 18:12:47 +01:00
Disty0 d4e2cbb826 SDNQ fix torch.compile always being active 2025-12-08 18:15:08 +03:00
Disty0 3ae7ecdbad SDNQ fix quantization_device getting ignored on post load quant 2025-12-08 01:29:52 +03:00
Disty0 064b64c76c cleanup 2025-12-08 01:14:19 +03:00
Disty0 6e05a12a49 SDNQ post process pre-quants after load 2025-12-08 01:08:53 +03:00
Disty0 0835ca6f66 SDNQ add explicit model.quantization_method = QuantizationMethod.SDNQ 2025-12-08 00:46:40 +03:00
Disty0 7a6356f8eb SDNQ fix transformers v5 and check for torch._dynamo.config.disable 2025-12-08 00:36:15 +03:00
Disty0 4f90054bf7 SDNQ transformers v5 support 2025-12-07 21:37:41 +03:00
Disty0 1cfb61809f cleanup 2025-12-05 18:40:49 +03:00
Disty0 5b86bef796 SDNQ add longcat keys 2025-12-05 18:37:20 +03:00
vladmandic 0ad40d2b8b lint
Signed-off-by: vladmandic <mandic00@live.com>
2025-12-02 12:25:04 +01:00
Disty0 7aa1bfdc70 Add get_modules_to_not_convert from transformers v5 2025-12-02 01:01:51 +03:00
Disty0 d9bc31e7da Cleanup 2025-11-29 01:46:04 +03:00
Disty0 01a0f6b356 Warn and disable quantized matmul if triton is not available 2025-11-29 01:34:54 +03:00