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
782 Commits 66 Branches 705 Tags
new_amdshark_design
Commit Graph

782 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Abhishek-Varma
ce00c1c5e1 [SharkInference] Make SharkInference compile the entire module 
-- Previously SharkInference was compiling and providing run APIs
   for a harcoded function with function name "forward".
-- This commit makes the compiling functionality generic and now
   any function being defined within the module can be run.
-- It also creates an API to fetch all the function names defined
   within the compiled module.

Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
 2022-12-24 09:05:06 +00:00
Stanley Winata
136021424c [SD] Change default VMA large heap block size for windows perf. (#715) 
Windows perform can boost from 2.67s/image to 2.4523s/image.
While Linux stays the same.
20221224.411 20221224.410 20221224.409 20221224.408 		2022-12-24 01:40:58 +07:00
PhaneeshB
fee4ba3746 Add openjourney 2022-12-23 23:34:22 +05:30
Gaurav Shukla
a5b70335d4 [SD][web] Add variant support in the web UI 
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-23 23:18:27 +05:30
Stanley Winata
5cf4976054 [Vulkan][utils] Add GTX Pascal support. (#709) 20221223.407 20221223.406 2022-12-22 15:24:15 -08:00
PhaneeshB
1aa3255061 Add vaebase for av3 and ad 2022-12-23 04:17:17 +05:30
Daniel Garvey
b01f29f10d add support for clear_all (#691) 2022-12-22 11:25:03 -06:00
Boian Petkantchin
2673abca88 Fix concurrency issue in stress_test for CUDA devices 2022-12-22 08:54:19 -08:00
Gaurav Shukla
7eeb7f0715 [SD] Update all the utilities to make web and CLI codebase closer (#707) 
At this point, all the utilities of SD web and CLI are exactly same.

Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>

Signed-off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-22 02:49:48 -08:00
powderluv
37262a2479 Remove spurious characters 20221222.405 20221222.404 2022-12-21 19:23:54 -08:00
Gaurav Shukla
de6e304959 [SD] Fix the resource location in shark_sd.spec (#706) 2022-12-21 14:41:56 -08:00
Quinn Dawkins
234475bbc7 Add base_vae entries for variant models (#705) 2022-12-21 14:35:08 -08:00
Quinn Dawkins
abbd9f7cfc [SD] Set unet flags for cuda (#704) 2022-12-21 13:22:04 -08:00
Gaurav Shukla
dfd6ba67b3 [SD] Update SD CLI to use model_db.json 
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-22 02:13:04 +05:30
yzhang93
1595254eab Modify model annotation tool to walk through ops by shape (#692) 2022-12-21 10:46:30 -08:00
PhaneeshB
6964c5eeba encapsulate relevant methods in one method 20221221.402 2022-12-21 23:56:17 +05:30
PhaneeshB
2befe771b3 Add support for automatic target triple selection for SD 2022-12-21 22:38:06 +05:30
Prashant Kumar
b133a035a4 Add the download progress bar. 2022-12-21 15:47:33 +05:30
Gaurav Shukla
726c062327 [SD] Update spec files 
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-21 14:16:04 +05:30
Gaurav Shukla
9083672de3 [SD][web] Tuned models only for stablediffusion/fp16 and rdna3 cards 
Currently tuned models are only available for stablediffusion/fp16 and
rdna3 cards.

Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-21 14:15:39 +05:30
Quinn Dawkins
cdbaf880af [SD] [web] Add model variants to web 2022-12-21 13:42:22 +05:30
Quinn Dawkins
9434981cdc Add random seed generation for seed = -1 in cli (#689) 2022-12-20 17:15:22 -05:00
Phaneesh Barwaria
8b3706f557 Add Anything v3 and AnalogDiffusion variants of SD (#685) 
* base support for anythingv3

* add analogdiffusiont

* Update readme

* keep max len 77 till support for 64 added for variants

* lint fix
 2022-12-20 13:08:13 -08:00
Gaurav Shukla
0d5173833d [SD] Add a json file for model names information. (#687) 
This commit simplifies the code to identify the model name for a
particular set of flags. This is achieved by introducing a json file
that stores the model names information. The models are uploaded in
gcloud with these names.

Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>

Signed-off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-20 11:47:31 -08:00
powderluv
bf1178eb79 roll to build 400 2022-12-20 10:34:31 -08:00
yzhang93
abcd3fa94a [SD] Set model max length 64 as default (#681) 20221220.400 2022-12-19 21:13:04 -08:00
Quinn Dawkins
62aa1614b6 [SD] Add --use_base_vae flag to do conversion to pixel space on cpu (#682) 2022-12-19 21:09:39 -08:00
Quinn Dawkins
7027356126 [SD] Fix warmup for max length 64 (#680) 2022-12-19 21:04:44 -05:00
yzhang93
5ebe13a13d Add Unet len 64 tuned model (#679) 2022-12-19 16:24:08 -08:00
Gaurav Shukla
c3bed9a2b7 [SD][web] Add flag to disable the progress bar animation 
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-20 02:50:04 +05:30
yzhang93
f865222882 Update VAE 19dec tuned model (#676) 2022-12-19 12:42:28 -08:00
powderluv
e2fe2e4095 Point to 398 2022-12-19 12:08:30 -08:00
powderluv
0532a95f08 Update stable_diffusion_amd.md 2022-12-19 12:04:42 -08:00
Quinn Dawkins
ff536f6015 [SD] Deduplicate initial noise generation (#677) 2022-12-19 14:38:41 -05:00
Gaurav Shukla
097d0f27bb [SD][web] Add 64 max_length support in SD web 
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-20 00:00:58 +05:30
Prashant Kumar
2257f87edf Update opt_params.py 2022-12-19 23:43:30 +05:30
PhaneeshB
a17800da00 Add 64 len f16 untuned mlir 2022-12-19 22:53:17 +05:30
Prashant Kumar
059c1b3a19 Disable vae --use_tuned version. 20221219.398 2022-12-19 22:45:45 +05:30
Stanley Winata
9a36816d27 [SD][CLI] Add a warmup phase (#670) 2022-12-20 00:14:23 +07:00
Gaurav Shukla
7986b9b20b [SD][WEB] Update VAE model and wrapper 
This commit updates VAE model which significantly improves performance
by an order of ~300ms.

Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-19 22:32:05 +05:30
Gaurav Shukla
b2b3a0a62b [SD] Move initial latent generation out of inference time 
The initial random latent generation is not taken into account
for total SD inference time.

Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
 2022-12-19 22:32:05 +05:30
Prashant Kumar
3173b7d1d9 Update VAE model and wrapper. 2022-12-19 19:54:50 +05:30
Gaurav Shukla
9d716d70d6 [SD][web] Fix performance issues on shark scheduler 
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
20221219.397 		2022-12-19 17:44:37 +05:30
Stanley Winata
e1901a8608 [SD][CL] Disable print at every iteration. (#664) 
Printing might incur extra time to runtime. Hence, we add a flag to hide it. To disable printing please set this flag `--hide_steps`.

Co-authored-by: Stanley <stanley@MacStudio.lan>
 2022-12-19 15:39:57 +07:00
Quinn Dawkins
7d0cbd8d90 [SD][web] Set default tuned unet to v2 (#663) 20221219.396 2022-12-19 11:50:08 +07:00
Quinn Dawkins
59358361f9 [SD] Make clip batch 2 for positive and negative prompts (#662) 
Combines the forward passes for each input prompt type into a single batched clip pass.
 2022-12-18 23:46:21 -05:00
Quinn Dawkins
7fea2d3b68 [SD] update default large heap size for web as well (#661) 20221219.395 2022-12-18 21:50:26 -05:00
Quinn Dawkins
b6d3ff26bd [SD] Change default VMA large heap block size (#660) 2022-12-18 21:41:46 -05:00
Stella Laurenzo
523e63f5c1 Fix NoneType exception if vulkan tuning flags not detected. (#659) 
(This goes on to produce compilation errors, but one step at a time)
 2022-12-18 16:40:56 -08:00
Stella Laurenzo
10630ab597 Add config stanza for NVIDIA RTX 2080. (#658) 
Just happened to have this card on my Windows machine and verified that the SD demo works on it.

```
Average step time: 144.26142692565918ms/it
Clip Inference Avg time (ms) = (205.001 + 44.000) / 2 = 124.501
VAE Inference time (ms): 281.001

Total image generation time: 7.856997728347778sec
```

I'd love to add an API upstream to derive compiler tuning flags from a host device.
 2022-12-18 16:40:47 -08:00