Files
spark-vllm-docker/mods/fix-qwen3-coder-next/fix_crash.diff
2026-02-12 15:56:32 -08:00

15 lines
712 B
Diff

diff --git a/vllm/v1/core/single_type_kv_cache_manager.py b/vllm/v1/core/single_type_kv_cache_manager.py
index 0b6b7ed42ac1..b6e0305a312d 100644
--- a/vllm/v1/core/single_type_kv_cache_manager.py
+++ b/vllm/v1/core/single_type_kv_cache_manager.py
@@ -1000,7 +1000,8 @@ def cache_blocks(self, request: Request, num_tokens: int) -> None:
for block in self.req_to_blocks[request.request_id][
num_cached_blocks_before:num_cached_blocks_after
]:
- assert block.block_hash is not None
+ if block.is_null:
+ continue
self.cached_blocks_this_step.add(block.block_hash)
def new_step_starts(self) -> None: