[3/N][warmup][DSv4] Migrate FA4 MLA and shared CuTeDSL kernels (#53565)
Signed-off-by: LopezCastroRoberto <rocastro@redhat.com> Co-authored-by: Codex <codex@openai.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
co-authored by
Codex
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent
1713b9866a
commit
42801b3a6b
@@ -35,7 +35,7 @@ from vllm.v1.kv_cache_interface import (
|
||||
from ..configs import InklingModelConfig
|
||||
from .layernorm import InklingRMSNorm
|
||||
from .ops.fa4_rel_attention import (
|
||||
INKLING_FA4_REL_ATTENTION_KERNEL,
|
||||
_INKLING_FA4_REL_ATTENTION_KERNEL,
|
||||
bucket_max_seqlen_q,
|
||||
inkling_fa4_num_splits,
|
||||
)
|
||||
@@ -174,6 +174,9 @@ class InklingAttention(nn.Module, AttentionLayerBase):
|
||||
compilation_config.static_forward_context[prefix] = self
|
||||
self.kv_cache = torch.tensor([]) # replaced by bind_kv_cache
|
||||
|
||||
if vllm_config.kernel_config.enable_jit_warmup:
|
||||
_INKLING_FA4_REL_ATTENTION_KERNEL.register_warmup()
|
||||
|
||||
def get_attn_backend(self) -> type[AttentionBackend]:
|
||||
return FlashAttentionBackend
|
||||
|
||||
@@ -289,7 +292,7 @@ class InklingAttention(nn.Module, AttentionLayerBase):
|
||||
num_kv_heads=self.num_kv_heads,
|
||||
max_kv_len=self._max_kv_len,
|
||||
)
|
||||
INKLING_FA4_REL_ATTENTION_KERNEL(
|
||||
_INKLING_FA4_REL_ATTENTION_KERNEL(
|
||||
q[:nt],
|
||||
key_cache,
|
||||
value_cache,
|
||||
|
||||
@@ -15,12 +15,12 @@ from .sconv import fused_sconv
|
||||
_LAZY_EXPORTS = {
|
||||
"silu_and_mul_triton": "silu_and_mul",
|
||||
"sink_silu_mul_epilogue": "silu_and_mul",
|
||||
"INKLING_FA4_REL_ATTENTION_KERNEL": "fa4_rel_attention",
|
||||
"_INKLING_FA4_REL_ATTENTION_KERNEL": "fa4_rel_attention",
|
||||
}
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from .fa4_rel_attention import ( # noqa: F401
|
||||
INKLING_FA4_REL_ATTENTION_KERNEL,
|
||||
_INKLING_FA4_REL_ATTENTION_KERNEL,
|
||||
)
|
||||
from .silu_and_mul import ( # noqa: F401
|
||||
silu_and_mul_triton,
|
||||
|
||||
@@ -433,4 +433,4 @@ class InklingFA4RelAttentionKernel(
|
||||
)
|
||||
|
||||
|
||||
INKLING_FA4_REL_ATTENTION_KERNEL = InklingFA4RelAttentionKernel()
|
||||
_INKLING_FA4_REL_ATTENTION_KERNEL = InklingFA4RelAttentionKernel()
|
||||
|
||||
Reference in New Issue
Block a user