[ModelRunnerV2][Hybrid model] Support kernel block size in hybrid model (#38831)

Signed-off-by: MengqingCao <cmq0113@163.com>
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Signed-off-by: Mengqing Cao <cmq0113@163.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>
This commit is contained in:
Mengqing Cao
2026-05-28 00:55:55 +00:00
committed by GitHub
co-authored by Nick Hill
parent 381edde1b9
commit 1223732dda
5 changed files with 82 additions and 75 deletions
+3 -4
View File
@@ -393,10 +393,9 @@ class GPUModelRunner(LoRAModelRunnerMixin):
) + spec.num_speculative_blocks
max_num_blocks_per_group.append(max_num_blocks)
(self.attn_backends, self.attn_groups, attn_cg_support, kernel_block_sizes) = (
init_attn_backend(self.kv_cache_config, self.vllm_config, self.device)
self.attn_groups, attn_cg_support, kernel_block_sizes = init_attn_backend(
self.kv_cache_config, self.vllm_config, self.device
)
self.block_tables = BlockTables(
block_sizes=block_sizes,
max_num_reqs=self.max_num_reqs,
@@ -440,7 +439,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
self.kv_caches,
self.compilation_config.static_forward_context,
self.kv_cache_config,
self.attn_backends,
self.attn_groups,
self.device,
self.cache_config.cache_dtype,
kernel_block_sizes,