patientx
06ac233007
Merge branch 'comfyanonymous:master' into master
2025-06-10 20:34:42 +03:00
comfyanonymous
6e28a46454
Apple most likely is never fixing the fp16 attention bug. ( #8485 )
2025-06-10 13:06:24 -04:00
patientx
4bc3866c67
Merge branch 'comfyanonymous:master' into master
2025-06-09 21:10:00 +03:00
comfyanonymous
7f800d04fa
Enable AMD fp8 and pytorch attention on some GPUs. ( #8474 )
...
Information is from the pytorch source code.
2025-06-09 12:50:39 -04:00
patientx
b4d015f5f3
Merge branch 'comfyanonymous:master' into master
2025-06-08 21:21:41 +03:00
comfyanonymous
97755eed46
Enable fp8 ops by default on gfx1201 ( #8464 )
2025-06-08 14:15:34 -04:00
patientx
156aedd995
Merge branch 'comfyanonymous:master' into master
2025-06-07 19:30:45 +03:00
comfyanonymous
daf9d25ee2
Cleaner torch version comparisons. ( #8453 )
2025-06-07 10:01:15 -04:00
patientx
07b8d211e6
Merge branch 'comfyanonymous:master' into master
2025-05-30 23:48:15 +03:00
comfyanonymous
704fc78854
Put ROCm version in tuple to make it easier to enable stuff based on it. ( #8348 )
2025-05-30 15:41:02 -04:00
patientx
8609a6dced
Merge branch 'comfyanonymous:master' into master
2025-05-27 01:03:35 +03:00
comfyanonymous
89a84e32d2
Disable initial GPU load when novram is used. ( #8294 )
2025-05-26 16:39:27 -04:00
patientx
bbcb33ea72
Merge branch 'comfyanonymous:master' into master
2025-05-26 16:26:39 +03:00
comfyanonymous
e5799c4899
Enable pytorch attention by default on AMD gfx1151 ( #8282 )
2025-05-26 04:29:25 -04:00
patientx
3b69a08c08
Merge branch 'comfyanonymous:master' into master
2025-05-24 04:06:28 +03:00
comfyanonymous
0b50d4c0db
Add argument to explicitly enable fp8 compute support. ( #8257 )
...
This can be used to test if your current GPU/pytorch version supports fp8 matrix mult in combination with --fast or the fp8_e4m3fn_fast dtype.
2025-05-23 17:43:50 -04:00
patientx
f49d26848b
Merge branch 'comfyanonymous:master' into master
2025-04-30 13:46:06 +03:00
comfyanonymous
0a66d4b0af
Per device stream counters for async offload. ( #7873 )
2025-04-29 20:28:52 -04:00
patientx
0aeb958ea5
Merge branch 'comfyanonymous:master' into master
2025-04-29 01:49:37 +03:00
comfyanonymous
5a50c3c7e5
Fix stream priority to support older pytorch. ( #7856 )
2025-04-28 13:07:21 -04:00
patientx
6244dfa1e1
Merge branch 'comfyanonymous:master' into master
2025-04-28 01:13:53 +03:00
comfyanonymous
c8cd7ad795
Use stream for casting if enabled. ( #7833 )
2025-04-27 05:38:11 -04:00
patientx
9cc8e2e1d0
Merge branch 'comfyanonymous:master' into master
2025-04-26 23:32:14 +03:00
comfyanonymous
0dcc75ca54
Add experimental --async-offload lowvram weight offloading. ( #7820 )
...
This should speed up the lowvram mode a bit. It currently is only enabled when --async-offload is used but it will be enabled by default in the future if there are no problems.
2025-04-26 16:11:21 -04:00
patientx
a397c3aeb3
Merge branch 'comfyanonymous:master' into master
2025-04-22 13:28:29 +03:00
comfyanonymous
2d6805ce57
Add option for using fp8_e8m0fnu for model weights. ( #7733 )
...
Seems to break every model I have tried but worth testing?
2025-04-22 06:17:38 -04:00
patientx
4541842b9a
Merge branch 'comfyanonymous:master' into master
2025-04-03 03:15:32 +03:00
BiologicalExplosion
2222cf67fd
MLU memory optimization ( #7470 )
...
Co-authored-by: huzhan <huzhan@cambricon.com>
2025-04-02 19:24:04 -04:00
patientx
1040220970
Merge branch 'comfyanonymous:master' into master
2025-04-01 22:56:01 +03:00
BVH
301e26b131
Add option to store TE in bf16 ( #7461 )
2025-04-01 13:48:53 -04:00
patientx
8115bdf68a
Merge branch 'comfyanonymous:master' into master
2025-03-25 22:35:14 +03:00
comfyanonymous
8edc1f44c1
Support more float8 types.
2025-03-25 05:23:49 -04:00
patientx
eaf40b802d
Merge branch 'comfyanonymous:master' into master
2025-03-14 12:00:25 +03:00
FeepingCreature
7aceb9f91c
Add --use-flash-attention flag. ( #7223 )
...
* Add --use-flash-attention flag.
This is useful on AMD systems, as FA builds are still 10% faster than Pytorch cross-attention.
2025-03-14 03:22:41 -04:00
comfyanonymous
35504e2f93
Fix.
2025-03-13 15:03:18 -04:00
comfyanonymous
299436cfed
Print mac version.
2025-03-13 10:05:40 -04:00
patientx
c469113159
Merge branch 'comfyanonymous:master' into master
2025-03-09 14:09:50 +03:00
comfyanonymous
0952569493
Fix stable cascade VAE on some lowvram machines.
2025-03-08 20:24:04 -05:00
patientx
0c4bebf5fb
Merge branch 'comfyanonymous:master' into master
2025-03-01 14:59:20 +03:00
Chenlei Hu
4d55f16ae8
Use enum list for --fast options ( #7024 )
2025-03-01 02:37:35 -05:00
patientx
af43425ab5
Update model_management.py
2025-02-28 16:37:55 +03:00
patientx
1871a594ba
Merge branch 'comfyanonymous:master' into master
2025-02-28 11:47:19 +03:00
comfyanonymous
cf0b549d48
--fast now takes a number as argument to indicate how fast you want it.
...
The idea is that you can indicate how much quality vs speed you want.
At the moment:
--fast 2 enables fp16 accumulation if your pytorch supports it.
--fast 5 enables fp8 matrix mult on fp8 models and the optimization above.
--fast without a number enables all optimizations.
2025-02-28 02:48:20 -05:00
comfyanonymous
eb4543474b
Use fp16 for intermediate for fp8 weights with --fast if supported.
2025-02-28 02:17:50 -05:00
comfyanonymous
1804397952
Use fp16 if checkpoint weights are fp16 and the model supports it.
2025-02-27 16:39:57 -05:00
patientx
c4fb9f2a63
Merge branch 'comfyanonymous:master' into master
2025-02-27 13:06:17 +03:00
BiologicalExplosion
89253e9fe5
Support Cambricon MLU ( #6964 )
...
Co-authored-by: huzhan <huzhan@cambricon.com>
2025-02-26 20:45:13 -05:00
patientx
d705fe2e0b
Merge branch 'comfyanonymous:master' into master
2025-02-24 13:42:27 +03:00
comfyanonymous
96d891cb94
Speedup on some models by not upcasting bfloat16 to float32 on mac.
2025-02-24 05:41:32 -05:00
patientx
8142770e5f
Merge branch 'comfyanonymous:master' into master
2025-02-23 14:51:43 +03:00
