Skip to content

Add cutile jsd #1228

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
vaibhavjindal merged 11 commits into from
May 27, 2026
Merged

Add cutile jsd #1228

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
8da96be
add cutile jsd example
xjmxyt May 18, 2026
f2593b2
refine
xjmxyt May 18, 2026
255137a
Merge branch 'main' into feature/add_cutile
vaibhavjindal May 18, 2026
7733ee4
refine readme
xjmxyt May 19, 2026
8ac80fd
Refactor backend registry: rename Vendor to Backend, support multi-de…
vaibhavjindal May 19, 2026
00e00d7
Merge branch 'main' into feature/add_cutile
xjmxyt May 21, 2026
edc14fd
Merge branch 'main' into feature/add_cutile
vaibhavjindal May 26, 2026
f51c8f7
Restructure: hardware backends in backends/, DSLs at top level
vaibhavjindal May 26, 2026
7777be5
Merge branch 'main' into feature/add_cutile
vaibhavjindal May 26, 2026
651911b
use a seperate csv file
xjmxyt May 27, 2026
2f049f9
Apply ruff format
vaibhavjindal May 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 19 additions & 1 deletion README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -139,6 +139,8 @@ pip3 install --pre torch torchvision torchaudio --index-url https://download.pyt
### Optional Dependencies

- `transformers >= 4.x`: Required if you plan to use the transformers models patching APIs. The specific model you are working will dictate the minimum version of transformers.
- `cuda-tile`: Required when enabling the optional cuTile backend on CUDA. Use this when your environment already provides CUDA Toolkit 13.1 or newer, or an existing tileiras compiler installation.
- `cuda-tile[tileiras]`: Required when enabling the optional cuTile backend with the tileiras compiler installed directly into your Python environment.

> **Note:**
> Our kernels inherit the full spectrum of hardware compatibility offered by [Triton](https://github.com/triton-lang/triton).
Expand Down Expand Up @@ -168,10 +170,26 @@ pip install -e .
# Setup Development Dependencies
pip install -e ".[dev]"

# Setup cuTile Dependencies
pip install -e ".[cutile]"

# Or install cuTile with the optional tileiras compiler
pip install -e ".[cutile-tileiras]"

# NOTE -> For AMD users only
pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm6.3/
```

### Enable cuTile Backend

cuTile is an optional CUDA-only DSL implementation. After installing the `cutile` or `cutile-tileiras` extra, enable it explicitly:

```bash
LIGER_KERNEL_IMPL=cutile python your_script.py
```

`LIGER_KERNEL_IMPL` selects an opt-in implementation registered with Liger (currently `cutile`). Selecting one on an unsupported device, or without the required dependencies installed, raises an error.


## Getting Started

Expand Down Expand Up @@ -290,7 +308,7 @@ loss.backward()
| **Kernel** | **API** |
|---------------------------------|-------------------------------------------------------------|
| RMSNorm | `liger_kernel.transformers.LigerRMSNorm` |
| Modulated RMSNorm | `liger_kernel.transformers.LigerModulatedRMSNorm` |
| Modulated RMSNorm | `liger_kernel.transformers.LigerModulatedRMSNorm` |
| LayerNorm | `liger_kernel.transformers.LigerLayerNorm` |
| RoPE | `liger_kernel.transformers.liger_rotary_pos_emb` |
| SwiGLU | `liger_kernel.transformers.LigerSwiGLUMLP` |
Expand Down
32 changes: 32 additions & 0 deletions benchmark/data/all_benchmark_data.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2187,3 +2187,35 @@ fused_moe,huggingface,backward,memory,MB,E,num_experts,16,2072.1728515625,2072.1
fused_moe,huggingface,backward,memory,MB,E,num_experts,32,2737.08349609375,2737.08349609375,2737.08349609375,"{""sweep_dim"": ""E"", ""T"": 8192, ""E"": null, ""H"": 2048, ""intermediate_dim"": 768, ""K"": 8, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2026-04-02 23:59:56,0.7.0
fused_moe,huggingface,backward,memory,MB,E,num_experts,64,4078.97021484375,4078.97021484375,4078.97021484375,"{""sweep_dim"": ""E"", ""T"": 8192, ""E"": null, ""H"": 2048, ""intermediate_dim"": 768, ""K"": 8, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2026-04-02 23:59:56,0.7.0
fused_moe,huggingface,backward,memory,MB,E,num_experts,128,6763.82275390625,6763.82275390625,6763.82275390625,"{""sweep_dim"": ""E"", ""T"": 8192, ""E"": null, ""H"": 2048, ""intermediate_dim"": 768, ""K"": 8, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2026-04-02 23:59:56,0.7.0
jsd,torch,full,speed,ms,BT,total tokens,1024,5.921823978424072,5.921823978424072,5.921823978424072,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:13:37,0.8.0
jsd,torch,full,speed,ms,BT,total tokens,2048,12.200063705444336,12.200063705444336,12.200063705444336,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:13:37,0.8.0
jsd,torch,full,speed,ms,BT,total tokens,4096,24.145984649658203,24.145984649658203,24.145984649658203,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:13:37,0.8.0
jsd,torch,full,speed,ms,BT,total tokens,8192,50.45283126831055,50.45283126831055,50.45283126831055,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:13:37,0.8.0
jsd,liger,full,speed,ms,BT,total tokens,1024,6.0959038734436035,6.0959038734436035,6.0959038734436035,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:28,0.8.0
jsd,liger,full,speed,ms,BT,total tokens,2048,10.940447807312012,10.940447807312012,10.940447807312012,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:28,0.8.0
jsd,liger,full,speed,ms,BT,total tokens,4096,21.781631469726562,21.781631469726562,21.781631469726562,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:28,0.8.0
jsd,liger,full,speed,ms,BT,total tokens,8192,44.07699203491211,44.07699203491211,44.07699203491211,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:28,0.8.0
jsd,torch,forward,speed,ms,BT,total tokens,1024,2.2900800704956055,2.2883904933929444,2.2906303882598875,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:29,0.8.0
jsd,torch,forward,speed,ms,BT,total tokens,2048,4.97105598449707,4.9135422706604,5.02856969833374,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:29,0.8.0
jsd,torch,forward,speed,ms,BT,total tokens,4096,9.907423973083496,9.907423973083496,9.907423973083496,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:29,0.8.0
jsd,torch,forward,speed,ms,BT,total tokens,8192,20.02751922607422,20.02751922607422,20.02751922607422,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:29,0.8.0
jsd,liger,forward,speed,ms,BT,total tokens,1024,5.783552169799805,5.783552169799805,5.783552169799805,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:29,0.8.0
jsd,liger,forward,speed,ms,BT,total tokens,2048,9.110560417175293,9.110560417175293,9.110560417175293,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:29,0.8.0
jsd,liger,forward,speed,ms,BT,total tokens,4096,18.322431564331055,18.322431564331055,18.322431564331055,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:29,0.8.0
jsd,liger,forward,speed,ms,BT,total tokens,8192,37.44358444213867,37.44358444213867,37.44358444213867,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:29,0.8.0
jsd,torch,backward,speed,ms,BT,total tokens,1024,3.7858558893203735,3.7852798938751224,3.786431884765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,torch,backward,speed,ms,BT,total tokens,2048,7.665791988372803,7.665791988372803,7.665791988372803,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,torch,backward,speed,ms,BT,total tokens,4096,15.20956802368164,15.20956802368164,15.20956802368164,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,torch,backward,speed,ms,BT,total tokens,8192,30.310592651367188,30.310592651367188,30.310592651367188,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,liger,backward,speed,ms,BT,total tokens,1024,1.0158560276031494,1.004588794708252,1.0225855827331543,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,liger,backward,speed,ms,BT,total tokens,2048,1.8555200099945068,1.8544960021972656,1.8571839809417723,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,liger,backward,speed,ms,BT,total tokens,4096,3.7145920991897583,3.7130560874938965,3.71612811088562,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,liger,backward,speed,ms,BT,total tokens,8192,7.243807792663574,7.243807792663574,7.243807792663574,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,torch,full,memory,MB,BT,total tokens,1024,6526.0009765625,6526.0009765625,6526.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,torch,full,memory,MB,BT,total tokens,2048,13026.0009765625,13026.0009765625,13026.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,torch,full,memory,MB,BT,total tokens,4096,26052.0,26052.0,26052.0,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,torch,full,memory,MB,BT,total tokens,8192,52104.0,52104.0,52104.0,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:30,0.8.0
jsd,liger,full,memory,MB,BT,total tokens,1024,3514.0009765625,3514.0009765625,3514.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:32,0.8.0
jsd,liger,full,memory,MB,BT,total tokens,2048,7014.0009765625,7014.0009765625,7014.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:32,0.8.0
jsd,liger,full,memory,MB,BT,total tokens,4096,14028.0009765625,14028.0009765625,14028.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:32,0.8.0
jsd,liger,full,memory,MB,BT,total tokens,8192,28056.0,28056.0,28056.0,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:14:32,0.8.0
33 changes: 33 additions & 0 deletions benchmark/data/all_benchmark_data_cutile.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,33 @@
kernel_name,kernel_provider,kernel_operation_mode,metric_name,metric_unit,x_name,x_label,x_value,y_value_50,y_value_20,y_value_80,extra_benchmark_config_str,gpu_name,timestamp,liger_version
jsd,torch,full,speed,ms,BT,total tokens,1024,5.9279680252075195,5.9279680252075195,5.9279680252075195,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:12,0.8.0
jsd,torch,full,speed,ms,BT,total tokens,2048,12.093536376953125,12.093536376953125,12.093536376953125,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:12,0.8.0
jsd,torch,full,speed,ms,BT,total tokens,4096,24.353023529052734,24.353023529052734,24.353023529052734,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:12,0.8.0
jsd,torch,full,speed,ms,BT,total tokens,8192,51.63132858276367,51.63132858276367,51.63132858276367,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:12,0.8.0
jsd,liger,full,speed,ms,BT,total tokens,1024,1.5985119938850403,1.5944639444351196,1.6005439758300781,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:15,0.8.0
jsd,liger,full,speed,ms,BT,total tokens,2048,3.0249600410461426,3.024307155609131,3.0514752864837646,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:15,0.8.0
jsd,liger,full,speed,ms,BT,total tokens,4096,6.043647766113281,6.043647766113281,6.043647766113281,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:15,0.8.0
jsd,liger,full,speed,ms,BT,total tokens,8192,12.18057632446289,12.18057632446289,12.18057632446289,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:15,0.8.0
jsd,torch,forward,speed,ms,BT,total tokens,1024,2.2989439964294434,2.2989439964294434,2.298969554901123,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,torch,forward,speed,ms,BT,total tokens,2048,4.600415945053101,4.598918342590332,4.60191354751587,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,torch,forward,speed,ms,BT,total tokens,4096,9.270400047302246,9.270400047302246,9.270400047302246,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,torch,forward,speed,ms,BT,total tokens,8192,19.314847946166992,19.314847946166992,19.314847946166992,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,liger,forward,speed,ms,BT,total tokens,1024,0.9553920030593872,0.9492863893508912,0.9575616240501403,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,liger,forward,speed,ms,BT,total tokens,2048,1.4541120529174805,1.4528576374053954,1.4553215980529786,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,liger,forward,speed,ms,BT,total tokens,4096,2.5651841163635254,2.5584064960479735,2.5675840854644774,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,liger,forward,speed,ms,BT,total tokens,8192,5.1241278648376465,5.1241278648376465,5.1241278648376465,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,torch,backward,speed,ms,BT,total tokens,1024,3.8217118978500366,3.8216639041900637,3.82175989151001,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,torch,backward,speed,ms,BT,total tokens,2048,7.542975902557373,7.542975902557373,7.542975902557373,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,torch,backward,speed,ms,BT,total tokens,4096,15.150239944458008,15.150239944458008,15.150239944458008,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,torch,backward,speed,ms,BT,total tokens,8192,30.65158462524414,30.65158462524414,30.65158462524414,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:16,0.8.0
jsd,liger,backward,speed,ms,BT,total tokens,1024,1.018943965435028,1.0006976008415223,1.0215808391571044,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:17,0.8.0
jsd,liger,backward,speed,ms,BT,total tokens,2048,1.8514400124549866,1.8510143756866455,1.8518656492233276,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:17,0.8.0
jsd,liger,backward,speed,ms,BT,total tokens,4096,3.6808160543441772,3.680499267578125,3.6811328411102293,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:17,0.8.0
jsd,liger,backward,speed,ms,BT,total tokens,8192,7.2151360511779785,7.2151360511779785,7.2151360511779785,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:17,0.8.0
jsd,torch,full,memory,MB,BT,total tokens,1024,6526.0009765625,6526.0009765625,6526.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:17,0.8.0
jsd,torch,full,memory,MB,BT,total tokens,2048,13026.0009765625,13026.0009765625,13026.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:17,0.8.0
jsd,torch,full,memory,MB,BT,total tokens,4096,26052.0,26052.0,26052.0,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:17,0.8.0
jsd,torch,full,memory,MB,BT,total tokens,8192,52104.0,52104.0,52104.0,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:17,0.8.0
jsd,liger,full,memory,MB,BT,total tokens,1024,3514.0009765625,3514.0009765625,3514.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:18,0.8.0
jsd,liger,full,memory,MB,BT,total tokens,2048,7014.0009765625,7014.0009765625,7014.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:18,0.8.0
jsd,liger,full,memory,MB,BT,total tokens,4096,14028.0009765625,14028.0009765625,14028.0009765625,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:18,0.8.0
jsd,liger,full,memory,MB,BT,total tokens,8192,28056.0,28056.0,28056.0,"{""vocab_size"": 128256, ""bsz"": 1, ""seq_len"": 8192}",NVIDIA B200,2026-05-27 17:16:18,0.8.0
4 changes: 3 additions & 1 deletion benchmark/scripts/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -452,7 +452,9 @@ def run_benchmarks(

print_benchmark_data(benchmark_data_list)

update_benchmark_data_csv(benchmark_data_list=benchmark_data_list, overwrite=overwrite)
impl_name = os.environ.get("LIGER_KERNEL_IMPL", "").strip().lower()
file_name = "all_benchmark_data.csv" if impl_name == "" else f"all_benchmark_data_{impl_name}.csv"
update_benchmark_data_csv(benchmark_data_list=benchmark_data_list, filename=file_name, overwrite=overwrite)


def parse_benchmark_script_args():
Expand Down
11 changes: 10 additions & 1 deletion setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,16 @@ def get_default_dependencies():

def get_optional_dependencies():
"""Get optional dependency groups."""
cutile_deps = [
"cuda-tile",
]
cutile_tileiras_deps = [
"cuda-tile[tileiras]",
]

return {
"cutile": cutile_deps,
"cutile-tileiras": cutile_tileiras_deps,
"dev": [
"transformers>=4.52.0",
"matplotlib>=3.7.2",
Expand All @@ -48,7 +57,7 @@ def get_optional_dependencies():
"mkdocs-material",
"torchvision>=0.20",
"prek>=0.2.28",
]
],
}


Expand Down
Loading
Loading