Skip to content

fix(grpo): use 64-bit row offsets in the GRPO loss kernels #192

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

Open
Lfan-ke wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix(grpo): use 64-bit row offsets in the GRPO loss kernels #192

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 1 addition & 1 deletion benchmarks/grpo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@

import torch
import triton
from dlblas.kernel.grpo_compute_loss_logits import GRPO_Loss_Optimized, grpo_compute_loss_torch
from dlblas.kernels.grpo_compute_loss_logits import GRPO_Loss_Optimized, grpo_compute_loss_torch

def run_benchmark():

Expand Down
6 changes: 4 additions & 2 deletions dlblas/kernels/grpo_compute_loss_logits.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,8 @@ def grpo_kernel(
HAS_DELTA: tl.constexpr, # 使用编译时常量来处理delta的有无
):

pid = tl.program_id(axis=0)
# 行偏移用 64 位：pid * V 在 int32 下会回绕，BL * V 超过 2^31 后越界
pid = tl.program_id(axis=0).to(tl.int64)
mask = tl.load(mask_ptr + pid)
if mask == 0:
tl.store(loss_i_ptr + pid, 0.0)
Expand Down Expand Up @@ -205,7 +206,8 @@ def grpo_bwd_kernel(
BLOCK_SIZE_V: tl.constexpr,
HAS_DELTA: tl.constexpr,
):
pid = tl.program_id(axis=0)
# 行偏移用 64 位：pid * V 在 int32 下会回绕，BL * V 超过 2^31 后越界
pid = tl.program_id(axis=0).to(tl.int64)
mask = tl.load(mask_ptr + pid)
if mask == 0:
grad_ptr = grad_new_logits_ptr + pid * V
Expand Down