[Kernel] Phase 4 batch-3: migrate tangled JIT subsystems + new groups into kernels.ops (RFC #29630) (#32045)
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 4.8
parent
71fe649d68
commit
74338e94f1
@@ -1132,11 +1132,14 @@ def dsa_impl_capability(impl: str) -> tuple[bool, str]:
|
||||
|
||||
if impl == "fa3":
|
||||
try:
|
||||
from sglang.jit_kernel.flash_attention import ( # noqa: F401
|
||||
from sglang.kernels.ops.attention.flash_attention import ( # noqa: F401
|
||||
flash_attn_with_kvcache,
|
||||
)
|
||||
except ImportError as exc:
|
||||
return False, f"sglang.jit_kernel.flash_attention unavailable: {exc}"
|
||||
return (
|
||||
False,
|
||||
f"sglang.kernels.ops.attention.flash_attention unavailable: {exc}",
|
||||
)
|
||||
# sgl-kernel flash_attn is compiled for SM9.x (Hopper) only;
|
||||
# it raises NotImplementedError on Blackwell (SM10.x+).
|
||||
if major < 9 or major >= 10:
|
||||
|
||||
@@ -6,8 +6,8 @@ from typing import List, Optional
|
||||
|
||||
import torch
|
||||
|
||||
from sglang.jit_kernel.kv_canary import consts
|
||||
from sglang.jit_kernel.kv_canary.verify import CANARY_SLOT_BYTES, RealKvSource
|
||||
from sglang.kernels.ops.kv_canary import consts
|
||||
from sglang.kernels.ops.kv_canary.verify import CANARY_SLOT_BYTES, RealKvSource
|
||||
from sglang.srt.kv_canary.buffer_group import CanaryBufferGroup, PoolKind
|
||||
from sglang.srt.kv_canary.config import CanaryConfig, CanaryMode
|
||||
from sglang.srt.kv_canary.pool_patcher.adapters.mha import attach_mha
|
||||
|
||||
@@ -5,8 +5,8 @@ from unittest.mock import patch
|
||||
|
||||
import torch
|
||||
|
||||
from sglang.jit_kernel.kv_canary.consts import RealKvHashMode
|
||||
from sglang.jit_kernel.kv_canary.verify import CanaryLaunchTag
|
||||
from sglang.kernels.ops.kv_canary.consts import RealKvHashMode
|
||||
from sglang.kernels.ops.kv_canary.verify import CanaryLaunchTag
|
||||
from sglang.srt.kv_canary import endpoint as endpoint_module
|
||||
from sglang.srt.kv_canary.buffer_group import CanaryBufferGroup
|
||||
from sglang.srt.kv_canary.capacities import CanaryLaunchCapacities
|
||||
|
||||
Reference in New Issue
Block a user