Skip to content

Revert "[MUSA][9/N] Add FA3 attention backend support through MATE (MUSA AI Tensor Engine)" #22002

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
Fridge003 merged 1 commit into from
Apr 3, 2026
Merged

Revert "[MUSA][9/N] Add FA3 attention backend support through MATE (MUSA AI Tensor Engine)" #22002

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
3 changes: 0 additions & 3 deletions python/pyproject_other.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,9 +116,6 @@ srt_musa = [
"torch_musa",
"torchada>=0.1.45",
"mthreads-ml-py",
"mate",
"mate-deep_gemm",
"mate-flash-attention",
"numpy<2.0",
]

Expand Down
6 changes: 0 additions & 6 deletions python/sglang/srt/configs/model_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -563,12 +563,6 @@ def _derive_model_shapes(self):
self.num_key_value_heads = getattr(
self.hf_text_config, "num_key_value_heads", None
)
self.first_k_dense_replace = getattr(
self.hf_text_config, "first_k_dense_replace", None
)
self.full_attention_interval = getattr(
self.hf_text_config, "full_attention_interval", None
)

# for Dbrx and MPT models
if self.hf_config.model_type in ["dbrx", "mpt"]:
Expand Down
3 changes: 0 additions & 3 deletions python/sglang/srt/environ.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -331,9 +331,6 @@ class Envs:
SGLANG_USE_AG_AFTER_QLORA = EnvBool(False)
SGLANG_NPU_FUSED_MOE_MODE = EnvInt(1)

# MTHREADS & MUSA
SGLANG_MUSA_FA3_FORCE_UPDATE_METADATA = EnvBool(False)

# Quantization
SGLANG_INT4_WEIGHT = EnvBool(False)
SGLANG_CPU_QUANTIZATION = EnvBool(False)
Expand Down
1 change: 0 additions & 1 deletion python/sglang/srt/hardware_backend/musa/__init__.py
View file
Open in desktop

This file was deleted.

14 changes: 0 additions & 14 deletions python/sglang/srt/hardware_backend/musa/attention/__init__.py
View file
Open in desktop

This file was deleted.

254 changes: 0 additions & 254 deletions python/sglang/srt/hardware_backend/musa/attention/flash_attention.py
View file
Open in desktop

This file was deleted.

23 changes: 7 additions & 16 deletions python/sglang/srt/layers/attention/attention_registry.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,6 @@
import logging
from typing import TYPE_CHECKING

from sglang.srt.utils import get_device_capability, is_musa

_is_musa = is_musa()

logger = logging.getLogger(__name__)


Expand Down Expand Up @@ -129,19 +125,14 @@ def create_flashmla_backend(runner):

@register_attention_backend("fa3")
def create_flashattention_v3_backend(runner):
import torch

major, minor = get_device_capability()
if not _is_musa:
assert (major == 8 and not runner.use_mla_backend) or major == 9, (
"FlashAttention v3 Backend requires SM>=80 and SM<=90. "
"Please use `--attention-backend flashinfer`."
)
else:
assert major >= 3 and minor >= 1, (
"FlashAttention v3 Backend requires MP>=31. "
"Please use `--attention-backend triton`."
)

assert (
torch.cuda.get_device_capability()[0] == 8 and not runner.use_mla_backend
) or torch.cuda.get_device_capability()[0] == 9, (
"FlashAttention v3 Backend requires SM>=80 and SM<=90. "
"Please use `--attention-backend flashinfer`."
)
from sglang.srt.layers.attention.flashattention_backend import FlashAttentionBackend

return FlashAttentionBackend(runner)
Expand Down
Loading
Loading