github/AMD-SHARK-Studio
1
1
Fork 0
mirror of https://github.com/nod-ai/AMD-SHARK-Studio.git synced 2026-04-03 03:00:17 -04:00
Code Issues Packages Projects Releases Wiki Activity
1,218 Commits 66 Branches 705 Tags
20230423.698
Commit Graph

1218 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
powderluv
d60a5a9396 Add fp32 for upscaler VAE (#1345) 20230423.698 2023-04-23 15:27:55 -07:00
m68k-fr
90df0ee365 [Web] Gallery set to a 768px reference for high-end desktop users (#1344) 2023-04-23 11:48:06 -07:00
nirvedhmeshram
133c1bcadd add device to scheduler model names (#1338) 20230422.697 2023-04-22 20:13:56 -05:00
powderluv
caadbe14e9 Revert VAE to use im2col (#1339) 20230422.696 2023-04-22 15:23:41 -07:00
Ean Garvey
5f5823ccd9 Fix inference object imports for SD apps. (#1334) 20230421.695 2023-04-21 13:40:48 -05:00
Vivek Khandelwal
d2f7e03b7e Add StableLM model (#1331) 2023-04-21 09:51:02 -07:00
Gaurav Shukla
0b01bbe479 [SD] Add txt2img/upscaler/inpaint/outpaint Rest API (#1325) 
Signed-off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2023-04-21 09:06:06 -07:00
yzhang93
25c5fc44ae Modify tuner.py to take vulkan target triple flag (#1328) 20230420.694 2023-04-20 14:31:32 -07:00
Daniel Garvey
7330729c92 enable sd pytest (#1322) 20230419.693 20230419.692 2023-04-19 22:11:30 -05:00
Ean Garvey
ce16cd5431 Create local shark_tank if needed for tuning configs. (#1321) 
Now that --clear_all successfully deletes local shark_tank cache, we need to make sure it exists before trying to use it.
 2023-04-19 11:44:21 -05:00
Ean Garvey
598dc5f79d Don't dump image data on img2img api call. (#1320) 2023-04-19 21:24:46 +05:30
Abhishek Varma
1f8e332cbe [SD] Fix img2img API bug for custom_vae argument (#1319) 
-- https://github.com/nod-ai/SHARK/pull/1314 misses to add `custom_vae`
   parameter to img2img_if's invocation within img2img_api.
-- This commit adds a fix to the same.

Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
Co-authored-by: Abhishek Varma <abhishek@nod-labs.com>
 2023-04-19 10:39:52 -05:00
Abhishek Varma
17b9632659 [SD] Adapted SHARK's v1 img2img API for SdPaint + updated Stencil model ID (#1318) 2023-04-19 06:29:36 -07:00
jinchen62
bda92a54ab Fix custom vae path (#1317) 20230418.690 2023-04-18 20:50:43 -07:00
jinchen62
747ed383b1 Add custom vae dropdown in webui (#1314) 20230418.684 2023-04-18 17:24:02 -07:00
Ean Garvey
1afe07c296 Disable winograd on VAE with rdna2 and fix unet tuning. (#1313) 
* Disable winograd on VAE with rdna2 and fix unet tuning.

* Fix batch size 1 downloads and clear_all on windows.
 2023-04-18 15:55:10 -05:00
jinchen62
b70919b38d Fix memory leak with ondemand (#1312) 
support ondemand for outpainting and multi batch_count
 2023-04-18 13:03:16 -05:00
m68k-fr
4e513d647f Update list of scheduler available for inferences (#1298) 2023-04-17 22:37:00 -05:00
jinchen62
94cd2a0fed Fix outpainting config (#1310) 2023-04-17 10:48:52 -07:00
Kyle Herndon
606029c01c Fix LoRA device format bug and allow LoRA to resume from a previous training 2023-04-17 13:19:46 +05:30
powderluv
1aa85222e9 Add AMD W7900 target triple (#1304) 
This maps to RDNA3
20230416.680 20230416.679 		2023-04-16 00:14:21 -07:00
m68k-fr
1b3f468c04 [Web] Style Fixes for Gradio V3.25.0 (#1300) 20230415.678 20230414.677 20230413.676 2023-04-13 18:40:42 -05:00
m68k-fr
35de7e27fa [Web] remove txt2img ui dependencies from png import metadata (#1275) 20230413.675 20230413.674 20230412.673 20230412.672 2023-04-12 07:32:47 -10:00
yzhang93
467f900759 Add auto-tuner to SD apps (#1291) 2023-04-12 09:21:17 -07:00
Ean Garvey
0bd9d582c7 Add documentation for using SHARK with AI-Render (#1296) 2023-04-12 03:09:34 -10:00
jinchen62
428cfe8dae Fix low vram mode issues (#1295) 
- add ondemand back to img2img
- workaround memory leak for batch count
20230411.671 		2023-04-11 17:59:09 -07:00
Ean Garvey
f17915bedc Fix batch size appending to model name. (#1294) 
* Update shark_downloader.py

* Update shark_downloader.py
 2023-04-11 15:34:25 -05:00
Gaurav Shukla
1b49b5149a [SD] Add Img2Img rest API 
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2023-04-11 23:06:58 +05:30
jinchen62
3002793301 Unload clip on demand and workaround memory leak (#1283) 20230410.670 2023-04-10 16:59:03 -07:00
Phaneesh Barwaria
d25ef5529f Add fix for vae fp32 Upscalar (#1284) 
- fixes size mismatch error for upscalar vae
20230409.669 20230408.668 20230407.667 		2023-04-07 14:36:40 -05:00
Ean Garvey
308856a947 Touch unet if base cfg needed for SD pipeline init (#1281) 20230406.666 20230405.665 20230405.664 2023-04-05 03:02:29 -05:00
m68k-fr
151b4e142f [SD] Fix encoder error for model_max_length not beeing 77 (#1278) 
Co-authored-by: powderluv <powderluv@users.noreply.github.com>
20230404.663 		2023-04-04 22:39:29 -07:00
Ean Garvey
e5a69a7c36 pin diffusers to e47459c (#1279) 20230404.662 20230404.661 20230404.660 2023-04-04 18:29:21 -07:00
m68k-fr
450b6cafc4 [SD] Add weight emphasis to prompts encoder (#1276) 20230404.659 20230404.658 2023-04-04 09:47:04 -07:00
Daniel Garvey
237d26baa2 update model db to reflect changes (#1277) 
* remove 1/1 tqdm progress bar

* update model_db to reflect changes
 2023-04-04 11:46:55 -05:00
Daniel Garvey
67d6ee1104 remove 1/1 tqdm progress bar (#1274) 20230403.657 2023-04-03 22:30:09 -05:00
Ean Garvey
98b069488e Add tank_version.json (#1272) 20230403.656 2023-04-03 18:36:23 -07:00
jinchen62
e0f227643a Fix webui circular import issue (#1271) 2023-04-03 16:00:10 -07:00
jinchen62
a0af3bb0cb xload and unload models (#1242) 20230403.655 2023-04-03 14:42:18 -07:00
powderluv
2cd61a5b96 strip source map (#1270) 2023-04-03 14:41:32 -07:00
Gaurav Shukla
f49d41a807 [SD] Add Stable diffusion text2image rest API (#1265) 
Signed-off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2023-04-03 12:02:24 -07:00
Ean Garvey
2191fc8952 Separate pytest benchmark modes and fix model updates for SHARK downloader / pytest. (#1264) 
* Only xfail windows models in CI

* downloader: make model updates more robust.

* Separate baseline and native benchmarks in pytest.

* Fix native benchmarks

* Fix torchvision model utils.
20230403.654 		2023-04-03 08:24:21 -07:00
PhaneeshB
aea7796e60 add gradio client to spec 20230403.653 2023-04-03 18:57:19 +05:30
Abhishek Varma
a376619f1e [SD] Improve vmfb caching algo and retry mechanism (#1248) 
-- This commit gets rid of the all-or-nothing vmfb caching mechanism
   and improves the retry mechanism by providing lower-level granularity
   for compiling each model units.

Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
Co-authored-by: Abhishek Varma <abhishek@nod-labs.com>
Co-authored-by: Ean Garvey <87458719+monorimet@users.noreply.github.com>
20230402.652 20230401.651 20230331.650 		2023-03-31 09:38:14 -07:00
powderluv
02d52bb626 Add Intel ARC A770 target triple (#1263) 
This just enables the plumbing. It generates black images.
20230330.649 20230329.648 		2023-03-29 14:49:05 -07:00
Abhishek Varma
3b63645f79 [SD] Fix custom model path for WebUI (#1260) 
-- This commit fixes custom model path for WebUI.

Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
Co-authored-by: Abhishek Varma <abhishek@nod-labs.com>
 2023-03-29 09:48:11 -07:00
Ean Garvey
d6f740b998 allow pytest to retry getting model artifacts + disable autotuning for pytorch benchmarks (#1257) 
* Adds a few xfails to enable macOS builder

* Convert string batch sizes to ints where needed.

* allow pytest to retry getting model artifacts

* Reduce attempts and add assert msg.
20230328.647 		2023-03-28 23:38:45 -05:00
Daniel Garvey
594c6b8ea2 fix ckpt dir (#1258) 20230328.646 2023-03-28 14:31:01 -07:00
Ean Garvey
96b1560da5 Make batch size configurable via pytest and fix sharktank generation. (#1227) 
* Fix sharktank generation and add batch_size pytest option for torch.

* Disable torch dynamo until py3.11 supported

* Compile torchmodel without dynamo if torch.compile fails

* Use release versions of TF/Keras for importer.

* Pin torchvision and remove debug prints.

* Remove duplicates from torch model list.

* Update generate_sharktank.py

* xfail a few models that fail sharktank generation/ numerics
 2023-03-28 14:33:39 -05:00
Abhishek Varma
0ef6a0e234 [SD] Fix Stencil scribble crash by updating image resize (#1255) 
-- This commit updates Stencil resize feature to cap the size of
   images within [128,768] as supported by the SD pipeline.
-- This solves the issue of scribble crashing on larger image.

Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
Co-authored-by: Abhishek Varma <abhishek@nod-labs.com>
20230328.643 		2023-03-28 10:13:11 -07:00