Disty0
|
878cab085f
|
Reverse the sdpa hijcak order
|
2025-02-14 19:56:39 +03:00 |
Disty0
|
f94196bcd1
|
Rename ROCm Flash atten hijack to CK Flash atten and enable AOTriton memory and flash atten by default
|
2025-02-13 22:01:06 +03:00 |
Vladimir Mandic
|
49712ab9e7
|
update requirements
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2025-02-13 12:03:08 -05:00 |
Vladimir Mandic
|
e28b8cd920
|
add torch gc debug
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2025-01-31 13:12:59 -05:00 |
Disty0
|
d2fee97e24
|
Update changelog
|
2025-01-31 20:27:22 +03:00 |
Disty0
|
039746914f
|
Add check for missing cuda and ipex params
|
2025-01-31 19:12:56 +03:00 |
Vladimir Mandic
|
3dcb70e8a2
|
device init logging
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2025-01-31 09:31:12 -05:00 |
Vladimir Mandic
|
1697fb1508
|
add tunable ops path
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2025-01-30 13:35:09 -05:00 |
Vladimir Mandic
|
0ea7840608
|
add tunable ops
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2025-01-30 13:08:49 -05:00 |
Disty0
|
a770b1c888
|
More correct Dynamic Atten SDPA implementation and deprecate IPEX Diffusers attention
|
2025-01-25 21:33:42 +03:00 |
Disty0
|
af35296a68
|
IPEX 4GB alloc detection and log driver version
|
2025-01-22 18:15:25 +03:00 |
Vladimir Mandic
|
bb97e695da
|
log cleanup
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-12-31 09:15:36 -05:00 |
Vladimir Mandic
|
910f5d0a73
|
lora direct on-demand apply/unapply
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-12-29 12:38:19 -05:00 |
Vladimir Mandic
|
20f2554cec
|
add sd35-ipadapter and more balanced offload optimizations
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-12-20 10:22:42 -05:00 |
Vladimir Mandic
|
e9f951b2c5
|
offload logging
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-12-11 14:20:01 -05:00 |
Vladimir Mandic
|
9a588d9c91
|
update balanced offload
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-12-11 12:06:03 -05:00 |
Vladimir Mandic
|
023b13b6cb
|
balanced offload improvements
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-12-01 15:34:25 -05:00 |
Vladimir Mandic
|
b7aff134a2
|
add low/high threshold to balanced offload
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-11-30 19:03:51 -05:00 |
Vladimir Mandic
|
b74166f9cb
|
detailer add augment setting
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-11-29 07:18:07 -05:00 |
Vladimir Mandic
|
dbb9ba0890
|
cuda memory limits
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-10-26 07:51:40 -04:00 |
Seunghoon Lee
|
a76893bd72
|
add cdna check
|
2024-10-26 14:09:50 +09:00 |
Seunghoon Lee
|
81bd236cc3
|
zluda&rocm bf16 test
|
2024-10-26 13:59:21 +09:00 |
Disty0
|
d459acfcca
|
Cleanup
|
2024-10-24 20:08:49 +03:00 |
Disty0
|
3b916d5e48
|
Zluda guess the GPU arch with the device name
|
2024-10-24 18:53:17 +03:00 |
Disty0
|
801ebdd080
|
Treat Zluda as a different backend and auto disable BF16 for Zluda and ROCm on RDNA1-2
|
2024-10-24 15:06:39 +03:00 |
Vladimir Mandic
|
0587f0be0c
|
gc logging
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-10-23 07:58:09 -04:00 |
Vladimir Mandic
|
64f363283f
|
messages,stats,save
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-10-20 09:26:07 -04:00 |
Disty0
|
6c11002420
|
PyTorch 2.5 XPU support
|
2024-10-17 23:11:52 +03:00 |
Disty0
|
b14e8f9a5f
|
Don't assume Cuda on devices.same_device()
|
2024-10-14 17:23:51 +03:00 |
Vladimir Mandic
|
c2ab0b11c3
|
check te device
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-10-14 09:29:33 -04:00 |
Disty0
|
9c05124c33
|
Add devices.has_xpu()
|
2024-10-13 15:13:38 +03:00 |
Disty0
|
011d9c3348
|
Move device backed initialization to shared.py
|
2024-10-13 14:56:28 +03:00 |
Disty0
|
84f8ab4076
|
Fix IPEX
|
2024-10-13 14:24:55 +03:00 |
Vladimir Mandic
|
0c54c235cb
|
add sageattention
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-10-12 15:43:34 -04:00 |
Disty0
|
2e2cb43406
|
Make SDPA hijacks chainable and add Sage Attention
|
2024-10-12 21:19:38 +03:00 |
Vladimir Mandic
|
ea0dfebe2d
|
better handle any quant lib requirements
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-10-12 13:36:16 -04:00 |
Vladimir Mandic
|
3bbcc33181
|
add detailer
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-10-07 09:32:56 -04:00 |
Vladimir Mandic
|
c21a10b7c9
|
add bf16 override for directml
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-09-29 21:34:59 -04:00 |
Vladimir Mandic
|
b31d02ba1d
|
cleanup
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-09-29 20:28:01 -04:00 |
Vladimir Mandic
|
47755dce6b
|
refactor devices
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-09-29 20:17:03 -04:00 |
Vladimir Mandic
|
fe94edf781
|
set default cuda dtype to auto
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-09-29 14:16:14 -04:00 |
Vladimir Mandic
|
174add0c3a
|
restore dtype after upcast complete
Signed-off-by: Vladimir Mandic <mandic00@live.com>
|
2024-09-29 13:22:08 -04:00 |
Vladimir Mandic
|
92f2a2902f
|
improve profiling
|
2024-09-23 11:07:24 -04:00 |
Vladimir Mandic
|
fe93ad6929
|
refactor xyzgrid
|
2024-09-13 19:22:02 -04:00 |
Disty0
|
586e5384b5
|
Update IPEX to 2.3 on Linux
|
2024-09-10 19:13:42 +03:00 |
Vladimir Mandic
|
b4df9a4de1
|
jumbo update with flux.1 refactor, see changelog for details
|
2024-09-01 22:56:15 -04:00 |
Vladimir Mandic
|
5ed58ac7cc
|
end-to-end update flux, see changelog and wiki
|
2024-08-28 08:04:24 -04:00 |
Disty0
|
a3f26c9df0
|
Convert Dynamic Attention SDP to a global SDP option
|
2024-08-18 01:44:27 +03:00 |
Disty0
|
f2769c0449
|
ROCm flash atten fall back to sdpa with fp32 inputs
|
2024-07-23 01:13:55 +03:00 |
xedis
|
809444eb5c
|
fix typo
|
2024-07-03 23:11:08 -07:00 |