Skip to content

Upgrade to torch 2.9.1 #1189

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

Closed
dxqb wants to merge 8 commits into from
Closed

Upgrade to torch 2.9.1 #1189

Show file tree
Hide file tree
Changes from 7 commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
68113b8
upgrade to torch 2.9.1
dxqb Dec 4, 2025
5d1d4ac
Merge branch 'upstream' into torch_2_9_1
dxqb Dec 15, 2025
1ab0c46
workaround for https://github.com/pytorch/pytorch/issues/170398
dxqb Dec 15, 2025
98ca133
Merge branch 'upstream' into torch_2_9_1
dxqb Dec 26, 2025
9af0046
merge
dxqb Dec 26, 2025
28e5d1b
Bump triton-windows ver. Uses shorter internal paths.
O-J1 Jan 6, 2026
2997d83
Bump B&B ver to 0.49.1 as discussed
O-J1 Jan 9, 2026
8fe4f0b
Bump av
O-J1 Jan 22, 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
4 changes: 2 additions & 2 deletions modules/util/triton_mm_8bit.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,7 @@
)

@triton.jit
def __mm_kernel(
def _mm_kernel(
a_ptr, b_ptr, c_ptr,
M, N, K,
stride_am, stride_ak,
Expand Down Expand Up @@ -109,7 +109,7 @@ def mm_8bit(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:

def grid(META):
return (triton.cdiv(N, META['BLOCK_SIZE_N']) , triton.cdiv(M, META['BLOCK_SIZE_M']), )
__mm_kernel[grid](
_mm_kernel[grid](
a, b, c,
M, N, K,
a.stride(0), a.stride(1),
Expand Down
12 changes: 6 additions & 6 deletions requirements-cuda.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@
# pytorch
--extra-index-url https://download.pytorch.org/whl/cu128
torch==2.8.0+cu128
torchvision==0.23.0+cu128
onnxruntime-gpu==1.22.0
nvidia-nccl-cu12==2.27.3; sys_platform == "linux"
triton-windows==3.4.0.post20; sys_platform == "win32"
torch==2.9.1+cu128
torchvision==0.24.1+cu128
onnxruntime-gpu==1.23.2
nvidia-nccl-cu12==2.27.5; sys_platform == "linux"
triton-windows==3.5.1.post24; sys_platform == "win32"

# optimizers
bitsandbytes==0.46.0 # bitsandbytes for 8-bit optimizers and weight quantization
bitsandbytes==0.49.1 # bitsandbytes for 8-bit optimizers and weight quantization
6 changes: 3 additions & 3 deletions requirements-default.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
# pytorch
torch==2.8.0
torchvision==0.23.0
onnxruntime==1.22.1
torch==2.9.1
torchvision==0.24.1
onnxruntime==1.23.2

# optimizers
# TODO
10 changes: 5 additions & 5 deletions requirements-global.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,16 +13,16 @@ yt-dlp #no pinned version, frequently updated for compatibility with sites
scenedetect==0.6.6

# pytorch
accelerate==1.7.0
safetensors==0.5.3
tensorboard==2.19.0
pytorch-lightning==2.5.1.post0
accelerate==1.12.0
safetensors==0.7.0
tensorboard==2.20.0
pytorch-lightning==2.6.0

# diffusion models
#Note: check whether Qwen bugs in diffusers have been fixed before upgrading diffusers (see BaseQwenSetup):
-e git+https://github.com/huggingface/diffusers.git@256e010#egg=diffusers
gguf==0.17.1
transformers==4.56.2
transformers==4.57.3
sentencepiece==0.2.1 # transitive dependency of transformers for tokenizer loading
omegaconf==2.3.0 # needed to load stable diffusion from single ckpt files
invisible-watermark==0.2.0 # needed for the SDXL pipeline
Expand Down
6 changes: 3 additions & 3 deletions requirements-rocm.txt
View file
Open in desktop
Copy link
Contributor

@hameerabbasi hameerabbasi Jan 9, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can add bitsandbytes here but not in requirements-global.txt or requirements-cpu.txt because macOS doesn't support 8-bit optimizers and that's like 90% of what we use bitsandbytes for.

Copy link
Collaborator

@O-J1 O-J1 Jan 9, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sounds good but that needs to be seperate (and tested!), I dont have a AMD gpu ;).

Original file line number Diff line number Diff line change
Expand Up @@ -3,9 +3,9 @@

# pytorch
--extra-index-url https://download.pytorch.org/whl/rocm6.3
torch==2.7.1+rocm6.3 #intentionally not upgraded because of reported problems
torchvision==0.22.1+rocm6.3
onnxruntime==1.22.1
torch==2.9.1+rocm6.3
torchvision==0.24.1+rocm6.3
onnxruntime==1.23.2

# optimizers
# TODO