[Perf][Kernel] Persistent TopK scheduler: unified CUDAGraph-safe kernel with dynamic per-row dispatch - DeepSeek-V3.2 DSA decode (#37421)

Signed-off-by: LopezCastroRoberto <rocastro@redhat.com>
Signed-off-by: Roberto L. Castro <38211239+LopezCastroRoberto@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
This commit is contained in:
Roberto L. Castro
2026-04-08 13:35:57 -04:00
committed by GitHub
co-authored by Claude Sonnet 4.5 Lucas Wilkinson
parent 75e01a39a1
commit b55d830ec7
9 changed files with 2050 additions and 494 deletions
+6 -2
View File
@@ -67,7 +67,9 @@ from vllm.model_executor.layers.quantization.utils.fp8_utils import (
per_token_group_quant_fp8,
)
from vllm.model_executor.layers.rotary_embedding import get_rope
from vllm.model_executor.layers.sparse_attn_indexer import SparseAttnIndexer
from vllm.model_executor.layers.sparse_attn_indexer import (
SparseAttnIndexer,
)
from vllm.model_executor.layers.vocab_parallel_embedding import (
ParallelLMHead,
VocabParallelEmbedding,
@@ -1203,7 +1205,9 @@ class DeepseekV2Model(nn.Module):
self.start_layer, self.end_layer, self.layers = make_layers(
config.num_hidden_layers,
lambda prefix: DeepseekV2DecoderLayer(
vllm_config, prefix, topk_indices_buffer=topk_indices_buffer
vllm_config,
prefix,
topk_indices_buffer=topk_indices_buffer,
),
prefix=f"{prefix}.layers",
)