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
1,743 Commits 66 Branches 705 Tags
diffusers-version
Commit Graph

622 Commits

Author SHA1 Message Date
Vivek Khandelwal
015d0132c3 Modify falcon pipeline to add fp16 support (#1551) 2023-06-19 09:57:13 -07:00
Chi_Liu
f60ab1f4fa Add Deberta to stablehlo in shark tank (#1545) 2023-06-16 13:24:44 -07:00
Surya Jasper
b203779462 Added Adreno target triples to vulkan_utils (#1543) 2023-06-15 16:42:59 -07:00
Ean Garvey
eb6d11cfed Change mlir dialects for tf tests to stablehlo. (#1535) 
* Change mlir dialects for tf tests to stablehlo

* Update shark_runner.py
 2023-06-14 10:43:49 -07:00
dan
bec82a665f mega vicuna merge 
single endpoint in apps/language/models/scripts/vicuna.py
removed main functions from pipelines
replaced divergent utils compile with shark_importer
adds support for different precisions
 2023-06-14 19:06:29 +05:30
Ean Garvey
9551490341 Remove deprecared --iree-mhlo-demote-164-to-132 flag usage. (#1533) 2023-06-13 22:40:47 -05:00
Ean Garvey
44273b0791 Fix conditional in transform_fx() (#1530) 2023-06-13 16:24:53 -05:00
Prashant Kumar
0a4c8fcb3e Minor changes in the fx transforms. 2023-06-13 21:23:35 +05:30
Stefan Kapusniak
2fec3c8169 re-indents add_upcast in shark importer (#1523) 
* The two with blocks in add_upcast appear to be underindented making
SD 1.4 break on rdna3, I've pushed them out one more tab, and then
everything appears to work again.
 2023-06-12 14:41:10 -05:00
Prashant Kumar
b6dbd20250 Modify the fx transforms. (#1521) 
- The bounds are set properly.
- The upcasting and downcasting is done for vicuna.
 2023-06-12 09:40:14 -07:00
Nithin Meganathan
34f1295349 Add a model config generator (#1511) 
Model config generator takes a PyTorch model as input and generates a JSON file with model layers and other propperties that define sharding on a particular hardware.
 2023-06-09 15:32:00 -07:00
Phaneesh Barwaria
1980d7b2c3 Cpu device map (#1515) 
* update cpu iree device

* fix vmfb paths vic unsharded
 2023-06-09 11:27:02 -05:00
Ean Garvey
a07fe80530 Update OPT, ResNet example scripts. (#1492) 
* Update API in OPT example.

* fix resnet50 script

* Add OPT1.3b test script.
 2023-06-05 20:19:35 -07:00
Ean Garvey
218ed78c40 Change instances of input_type='mhlo' to 'auto' (#1482) 2023-06-02 16:43:47 -05:00
Foxlum
5915bf7de3 Add to and tweak vulkan configuration environments. (#1475) 
* Update vulkan_target_env_utils.py

* Update vulkan_target_env_utils.py

Adjust target environment capabilities.

* Update vulkan_target_env_utils.py

black linted?
 2023-06-01 22:25:20 -07:00
yzhang93
47b346d428 Modify the lowering config format for SPIRVMatmulPromoteVectorize pipeline (#1471) 2023-05-29 21:53:48 -07:00
Ean Garvey
54e57f7771 Revive SD downloads from shark_tank. (#1465) 2023-05-25 12:03:21 -05:00
Prashant Kumar
8e571d165f Enable cpu f16 dtype tracing for the vicuna model. (#1461) 2023-05-24 09:37:57 -07:00
Chi_Liu
64c2b2d96b Add gpt2 to stablehlo support in shark tank (#1447) 
- Add torch decomposition support when generating shark tank
- Add gpt2 stablehlo
 2023-05-22 10:45:51 -07:00
Boian Petkantchin
20114deea0 In MiniLM JAX example verify MLIR result against JAX 2023-05-16 09:54:07 -07:00
Boian Petkantchin
bdf37b5311 If device/backend is unknown pass it to IREE verbatim 2023-05-16 09:54:07 -07:00
Chi_Liu
a0097a1ead Add mlir_type for torch_model_list.csv (#1428) 
- Enable stablehlo/tosa mlir output for torch model
- Add BERT stablehlo support
 2023-05-15 10:23:54 -07:00
Ean Garvey
a9bae00606 Fix vulkan device selection at compile time and adapt to IREE python changes. (#1407) 
* Add support for vulkan device selection at compile time.

* Don't convert device ID to int and fix .exe imports
 2023-05-12 23:31:50 -07:00
Ean Garvey
6c95ac0f37 Revert dialect registration in model annotator (#1406) 
Matches https://github.com/nod-ai/SHARK-Runtime/pull/58
 2023-05-09 11:50:19 -07:00
powderluv
d816cc015e Revert "added standalone vicuna script (#1399)" (#1402) 
This reverts commit 0e4a8ca240.
 2023-05-05 16:08:05 -07:00
Eliasj42
0e4a8ca240 added standalone vicuna script (#1399) 
Co-authored-by: Elias Joseph <elias@nod-labs.com>
 2023-05-05 15:46:05 -07:00
Ean Garvey
cdf2d61d53 Remove imports from iree.compiler.transforms from model annotator. (#1392) 2023-05-04 20:40:19 -07:00
Boian Petkantchin
eba4d06405 In MiniLM JAX example do not hardcode device (#1385) 
* In MiniLM JAX example do not hardcode device

* In MiniLM JAX example don't use bytecode MLIR

---------

Co-authored-by: Boian Petkantchin <boian@nod-labs.com>
 2023-05-03 10:34:42 -07:00
Boian Petkantchin
4cfba153d2 Add example JAX MiniLM inference (#1380) 
* Do not hardcode the name of the VM module in get_iree_module

* Add example JAX MiniLM inference

---------

Co-authored-by: Boian Petkantchin <boian@nod-labs.com>
 2023-05-02 15:03:54 -07:00
Chi_Liu
aa8ada9da9 Add support for torch to stablehlo and tosa in shark_importer (#1360) 2023-04-27 08:09:45 -07:00
powderluv
09e3f63d5b Fix pascal (#1346) 
* Add fp32 for upscaler VAE

* Plumb Pascal vulkan support
 2023-04-23 20:28:25 -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
powderluv
1aa85222e9 Add AMD W7900 target triple (#1304) 
This maps to RDNA3
 2023-04-16 00:14:21 -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
Ean Garvey
98b069488e Add tank_version.json (#1272) 2023-04-03 18:36:23 -07:00
powderluv
2cd61a5b96 strip source map (#1270) 2023-04-03 14:41:32 -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.
 2023-04-03 08:24:21 -07:00
powderluv
02d52bb626 Add Intel ARC A770 target triple (#1263) 
This just enables the plumbing. It generates black images.
 2023-03-29 14:49:05 -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.
 2023-03-28 23:38:45 -05:00
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
Thomas
807947c0c8 Remove deprecated cli option iree-hal-cuda-disable-loop-nounroll-wa (#1235) 2023-03-22 12:05:15 -05:00
Ean Garvey
d31bae488c Set iree-input-type to tm_tensor for SD (#1228) 2023-03-21 19:07:31 -07:00
gpetters94
7899e1803a Add fix for attention slicing fp16 (#1217) 2023-03-20 19:11:29 -07:00
Phoenix Meadowlark
d319f4684e Add peak memory reporting for IREE, TF and PyTorch (#1216) 2023-03-20 15:40:49 -05:00
Ean Garvey
54d7b6d83e Generate model artifacts in pytests if they don't exist in the cloud. (#1121) 
* Add gen_shark_files fn to shark_downloader for OTF artifact generation

* add generate_sharktank as a tank/ python module.

* Fix some paths in tank generation.
 2023-03-20 12:13:19 -05:00
Daniel Garvey
bdbe992769 Add IREE_SAVE_TEMPS for import_debug command (#1184) 
based on hf_model_id. Works on windows
 2023-03-14 11:40:23 -07:00
Boian Petkantchin
a8ce7680db Add flag to augment the device allocator (#1182) 
Example:
$ python my_app.py --device_allocator caching debug
This will wrap the device allocator with first caching allocator then
debug allocator.

$ python my_app.py --device_allocator caching
Only wrap with caching allocator.

Co-authored-by: Boian Petkantchin <boian@nod-labs.com>
 2023-03-13 15:49:26 -07:00
Daniel Garvey
62b5a9fd49 generate sharktank for apps dir (#966) 
* merge confix resolution

* add support to other scripts

---------

Co-authored-by: dan <dan@nod-labs.com>
 2023-03-13 10:54:15 -07:00
powderluv
f898a1d332 Update README.md 2023-03-12 16:54:42 -07:00
Prashant Kumar
97fdff7f19 Add instructions how to run the LLaMA model. (#1168) 
* Add instructions how to run the LLaMA model.

* Update README.md
 2023-03-10 12:36:37 -08:00