| Commit message (Collapse) | Author | Age | Files | Lines | |
|---|---|---|---|---|---|
| * | Make sub-quadratic the default for MPS | brkirch | 2023-08-13 | 1 | -2/+5 |
| | | |||||
| * | Use fixed size for sub-quadratic chunking on MPS | brkirch | 2023-08-13 | 1 | -1/+5 |
| | | | | | Even if this causes chunks to be much smaller, performance isn't significantly impacted. This will usually reduce memory usage but should also help with poor performance when free memory is low. | ||||
| * | update doggettx cross attention optimization to not use an unreasonable ↵ | AUTOMATIC1111 | 2023-08-02 | 1 | -2/+2 |
| | | | | | amount of memory in some edge cases -- suggestion by MorkTheOrk | ||||
| * | get attention optimizations to work | AUTOMATIC1111 | 2023-07-13 | 1 | -7/+7 |
| | | |||||
| * | SDXL support | AUTOMATIC1111 | 2023-07-12 | 1 | -8/+43 |
| | | |||||
| * | Merge pull request #11066 from aljungberg/patch-1 | AUTOMATIC1111 | 2023-06-07 | 1 | -1/+1 |
| |\ | | | | | Fix upcast attention dtype error. | ||||
| | * | Fix upcast attention dtype error. | Alexander Ljungberg | 2023-06-06 | 1 | -1/+1 |
| | | | | | | | | | | | | | | | | | | | | | | | Without this fix, enabling the "Upcast cross attention layer to float32" option while also using `--opt-sdp-attention` breaks generation with an error: ``` File "/ext3/automatic1111/stable-diffusion-webui/modules/sd_hijack_optimizations.py", line 612, in sdp_attnblock_forward out = torch.nn.functional.scaled_dot_product_attention(q, k, v, dropout_p=0.0, is_causal=False) RuntimeError: Expected query, key, and value to have the same dtype, but got query.dtype: float key.dtype: float and value.dtype: c10::Half instead. ``` The fix is to make sure to upcast the value tensor too. | ||||
| * | | Merge pull request #10990 from vkage/sd_hijack_optimizations_bugfix | AUTOMATIC1111 | 2023-06-04 | 1 | -1/+1 |
| |\ \ | | | | | | | torch.cuda.is_available() check for SdOptimizationXformers | ||||
| | * | | fix the broken line for #10990 | AUTOMATIC | 2023-06-04 | 1 | -1/+1 |
| | | | | |||||
| | * | | torch.cuda.is_available() check for SdOptimizationXformers | Vivek K. Vasishtha | 2023-06-03 | 1 | -1/+1 |
| | |/ | |||||
| | * | revert default cross attention optimization to Doggettx | AUTOMATIC | 2023-06-01 | 1 | -3/+3 |
| | | | | | | | | | make --disable-opt-split-attention command line option work again | ||||
| * | | revert default cross attention optimization to Doggettx | AUTOMATIC | 2023-06-01 | 1 | -3/+3 |
| | | | | | | | | | make --disable-opt-split-attention command line option work again | ||||
| * | | rename print_error to report, use it with together with package name | AUTOMATIC | 2023-05-31 | 1 | -2/+1 |
| | | | |||||
| * | | Add & use modules.errors.print_error where currently printing exception info ↵ | Aarni Koskela | 2023-05-29 | 1 | -4/+2 |
| |/ | | | | by hand | ||||
| * | Add a couple `from __future__ import annotations`es for Py3.9 compat | Aarni Koskela | 2023-05-20 | 1 | -0/+1 |
| | | |||||
| * | Apply suggestions from code review | AUTOMATIC1111 | 2023-05-19 | 1 | -38/+28 |
| | | | | Co-authored-by: Aarni Koskela <akx@iki.fi> | ||||
| * | fix linter issues | AUTOMATIC | 2023-05-18 | 1 | -1/+1 |
| | | |||||
| * | make it possible for scripts to add cross attention optimizations | AUTOMATIC | 2023-05-18 | 1 | -3/+132 |
| | | | | | add UI selection for cross attention optimization | ||||
| * | Autofix Ruff W (not W605) (mostly whitespace) | Aarni Koskela | 2023-05-11 | 1 | -16/+16 |
| | | |||||
| * | ruff auto fixes | AUTOMATIC | 2023-05-10 | 1 | -7/+7 |
| | | |||||
| * | autofixes from ruff | AUTOMATIC | 2023-05-10 | 1 | -1/+0 |
| | | |||||
| * | Fix for Unet NaNs | brkirch | 2023-05-08 | 1 | -0/+3 |
| | | |||||
| * | Update sd_hijack_optimizations.py | FNSpd | 2023-03-24 | 1 | -1/+1 |
| | | |||||
| * | Update sd_hijack_optimizations.py | FNSpd | 2023-03-21 | 1 | -1/+1 |
| | | |||||
| * | sdp_attnblock_forward hijack | Pam | 2023-03-10 | 1 | -0/+24 |
| | | |||||
| * | argument to disable memory efficient for sdp | Pam | 2023-03-10 | 1 | -0/+4 |
| | | |||||
| * | scaled dot product attention | Pam | 2023-03-06 | 1 | -0/+42 |
| | | |||||
| * | Add UI setting for upcasting attention to float32 | brkirch | 2023-01-25 | 1 | -60/+99 |
| | | | | | | | Adds "Upcast cross attention layer to float32" option in Stable Diffusion settings. This allows for generating images using SD 2.1 models without --no-half or xFormers. In order to make upcasting cross attention layer optimizations possible it is necessary to indent several sections of code in sd_hijack_optimizations.py so that a context manager can be used to disable autocast. Also, even though Stable Diffusion (and Diffusers) only upcast q and k, unfortunately my findings were that most of the cross attention layer optimizations could not function unless v is upcast also. | ||||
| * | better support for xformers flash attention on older versions of torch | AUTOMATIC | 2023-01-23 | 1 | -24/+18 |
| | | |||||
| * | add --xformers-flash-attention option & impl | Takuma Mori | 2023-01-21 | 1 | -2/+24 |
| | | |||||
| * | extra networks UI | AUTOMATIC | 2023-01-21 | 1 | -5/+5 |
| | | | | | rework of hypernets: rather than via settings, hypernets are added directly to prompt as <hypernet:name:weight> | ||||
| * | Added license | brkirch | 2023-01-06 | 1 | -0/+1 |
| | | |||||
| * | Change sub-quad chunk threshold to use percentage | brkirch | 2023-01-06 | 1 | -9/+9 |
| | | |||||
| * | Add Birch-san's sub-quadratic attention implementation | brkirch | 2023-01-06 | 1 | -25/+99 |
| | | |||||
| * | Use other MPS optimization for large q.shape[0] * q.shape[1] | brkirch | 2022-12-21 | 1 | -4/+6 |
| | | | | | | | Check if q.shape[0] * q.shape[1] is 2**18 or larger and use the lower memory usage MPS optimization if it is. This should prevent most crashes that were occurring at certain resolutions (e.g. 1024x1024, 2048x512, 512x2048). Also included is a change to check slice_size and prevent it from being divisible by 4096 which also results in a crash. Otherwise a crash can occur at 1024x512 or 512x1024 resolution. | ||||
| * | cleanup some unneeded imports for hijack files | AUTOMATIC | 2022-12-10 | 1 | -3/+0 |
| | | |||||
| * | do not replace entire unet for the resolution hack | AUTOMATIC | 2022-12-10 | 1 | -28/+0 |
| | | |||||
| * | Patch UNet Forward to support resolutions that are not multiples of 64 | Billy Cao | 2022-11-23 | 1 | -0/+31 |
| | | | | | Also modifed the UI to no longer step in 64 | ||||
| * | Remove wrong self reference in CUDA support for invokeai | Cheka | 2022-10-19 | 1 | -1/+1 |
| | | |||||
| * | Update sd_hijack_optimizations.py | C43H66N12O12S2 | 2022-10-18 | 1 | -0/+3 |
| | | |||||
| * | readd xformers attnblock | C43H66N12O12S2 | 2022-10-18 | 1 | -0/+15 |
| | | |||||
| * | delete xformers attnblock | C43H66N12O12S2 | 2022-10-18 | 1 | -12/+0 |
| | | |||||
| * | Use apply_hypernetwork function | brkirch | 2022-10-11 | 1 | -10/+4 |
| | | |||||
| * | Add InvokeAI and lstein to credits, add back CUDA support | brkirch | 2022-10-11 | 1 | -0/+13 |
| | | |||||
| * | Add check for psutil | brkirch | 2022-10-11 | 1 | -4/+15 |
| | | |||||
| * | Add cross-attention optimization from InvokeAI | brkirch | 2022-10-11 | 1 | -0/+79 |
| | | | | | | | * Add cross-attention optimization from InvokeAI (~30% speed improvement on MPS) * Add command line option for it * Make it default when CUDA is unavailable | ||||
| * | rename hypernetwork dir to hypernetworks to prevent clash with an old ↵ | AUTOMATIC | 2022-10-11 | 1 | -1/+1 |
| | | | | | filename that people who use zip instead of git clone will have | ||||
| * | fixes related to merge | AUTOMATIC | 2022-10-11 | 1 | -1/+2 |
| | | |||||
| * | replace duplicate code with a function | AUTOMATIC | 2022-10-11 | 1 | -29/+15 |
| | | |||||
| * | remove functorch | C43H66N12O12S2 | 2022-10-10 | 1 | -2/+0 |
| | | |||||
 |
index : stable-diffusion-webui-gfx803.git | |
| stable-diffusion-webui by AUTOMATIC1111 with patches for gfx803 GPU and Dockerfile |
| aboutsummaryrefslogtreecommitdiffstats |