Skip to content

Enable Triton MXFP4 MoE on gfx950 for GPT-OSS #220

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
ChuanLi1101 wants to merge 1 commit into
base: gpt-oss-allreduce-rmsnorm-fusion
Choose a base branch
from
Open

Enable Triton MXFP4 MoE on gfx950 for GPT-OSS #220

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
6 changes: 5 additions & 1 deletion atom/model_ops/moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@
per_tensor_dequantize,
shuffle_weights,
)
from atom.utils import envs
from atom.utils.custom_register import direct_register_custom_op
from atom.utils.forward_context import get_forward_context
from torch import nn
Expand Down Expand Up @@ -634,7 +635,10 @@ def __init__(self, quant_config: QuantizationConfig, moe: FusedMoEConfig):
self.quant_type == QuantType.per_1x128
or self.quant_type == QuantType.per_1x32
)
self.use_triton = get_gfx().startswith("gfx94")
gfx = get_gfx()
self.use_triton = gfx.startswith("gfx94") or (
gfx.startswith("gfx95") and envs.ATOM_USE_TRITON_GEMM
)
if self.use_triton:
from atom.model_ops.utils import has_triton_kernels

Expand Down