biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CI] Revert PRs 34818 and 33600 (#34979)

Browse Source
This commit is contained in:
Lucas Wilkinson
2026-02-20 16:25:50 -05:00
committed by GitHub
parent f24b2de3d3
commit aaefc58ee0
16 changed files with 249 additions and 301 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/v1/engine/core.py
View File

@@ -114,14 +114,7 @@ class EngineCore:
num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
vllm_config
)
if kv_cache_config.kv_cache_groups:
vllm_config.cache_config.block_size = min(
g.kv_cache_spec.block_size for g in kv_cache_config.kv_cache_groups
)
elif vllm_config.cache_config.block_size is None:
# Attention-free models (encoder-only, SSM) — use default.
vllm_config.cache_config.block_size = 16
vllm_config.validate_block_size()
vllm_config.cache_config.num_gpu_blocks = num_gpu_blocks
vllm_config.cache_config.num_cpu_blocks = num_cpu_blocks
self.collective_rpc("initialize_cache", args=(num_gpu_blocks, num_cpu_blocks))