AI/sdnext
1
0
Fork 0
mirror of https://github.com/vladmandic/sdnext.git synced 2026-01-27 15:02:48 +03:00
Code Activity
12,229 Commits 3 Branches 28 Tags
dev
Commit Graph

28 Commits

Author SHA1 Message Date
Seunghoon Lee
6f7fe5e7cc fix directml generator. 2024-09-20 10:09:00 +09:00
Disty0
e3b087b6c0 Add balanced offload mode and make offload modes a single choice list 2024-08-11 17:27:30 +03:00
Seunghoon Lee
ff2c1db1cc fix code formatting under modules/dml 2024-02-05 22:43:10 +09:00
Vladimir Mandic
1de3c6af0f add amused model support and update reference list 2023-12-27 10:15:36 -05:00
Vladimir Mandic
564d04d9f4 Merge pull request #2296 from vladmandic/master 
update dml
 2023-10-08 07:26:55 -04:00
Seunghoon Lee
e5f8b7f0a4 update dml 2023-10-08 02:19:56 +09:00
Vladimir Mandic
0afcfe6097 logger early init 2023-09-23 23:44:34 -04:00
Vladimir Mandic
496bdf7c55 fix backend switch 2023-09-15 08:39:25 -04:00
Vladimir Mandic
8fd96d0f30 catch directml and ipex initialization errors 2023-09-07 07:27:54 -04:00
Hameer Abbasi
4c7792ed01 Add pre-commit. 2023-09-04 13:57:47 +02:00
Seunghoon Lee
d711880aa9 New option for DirectML: memory stats provider. 
1. Performance Counter.
    Get vram size allocated to & used by python.exe from pdh.dll.
    Generation can be slower than atiadlxx.
    Use memory less greedy then atiadlxx.
    Windows only.
2. atiadlxx.
    Get max vram size and available vram size from AMD GPU driver (atiadlxx.dll).
    Use memory more greedy than Performance Counter.
    Windows & WSL are supported.
3. None.
    Assume available vram size is 8GB.
    Use memory regardless of current vram usage.
 2023-08-01 01:58:04 +09:00
Seunghoon Lee
42c6147ac8 cleanup 2023-07-29 13:48:28 +09:00
Seunghoon Lee
47f2f50574 Restore Python 3.9 compatibility. (DirectML) 2023-07-29 12:08:22 +09:00
Seunghoon Lee
77de9cd093 Fix medvram with DirectML. 2023-07-28 23:18:28 +09:00
Seunghoon Lee
0f44332e5c Make sequential CPU offload available for non-CUDA 
Add settings override for DirectML.
Move `devices.set_cuda_params()` to correct line.
 2023-07-28 23:11:57 +09:00
Seunghoon Lee
d4aa840a77 DirectML update. 
DirectML reuses GPU memory instead of returning it.
So prints "practical" GPU memory utilization too.
 2023-07-24 16:10:01 +09:00
Seunghoon Lee
0a52c44e73 DirectML rework & provide GPU memory usage (AMD only). 2023-07-15 18:55:38 +09:00
Seunghoon Lee
578f206bf7 DirectML rework. 2023-07-15 16:19:38 +09:00
Vladimir Mandic
2a21196061 Merge branch 'master' into dev 2023-07-08 13:35:25 -04:00
Seunghoon Lee
99c7f23c89 remove useless methods. 2023-07-09 01:29:04 +09:00
Seunghoon Lee
993de932ab Add an opts override for DirectML. 2023-07-06 01:35:00 +09:00
Vladimir Mandic
cb307399dd jumbo merge 2023-06-13 11:59:56 -04:00
Vladimir Mandic
5134471bc8 dml autocast 2023-05-14 13:24:59 -04:00
Vladimir Mandic
c46f0dbdeb fix 2023-05-13 12:55:50 -04:00
Vladimir Mandic
d96ab6a1ae update directml 2023-05-13 11:21:11 -04:00
Seunghoon Lee
d2d5011bd3 Implement memory estimation for AMDGPUs. 
Stable.
 2023-04-26 17:44:32 +09:00
Seunghoon Lee
df0e89be48 fix. 
Unstable & need more test.
 2023-04-26 12:45:44 +09:00
Seunghoon Lee
09ae33cdf7 Implement torch.dml. 
VERY UNSTABLE & NOT TESTED.
 2023-04-26 12:21:44 +09:00