[Model Runner v2] Fix block table IMA issue (#40648)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>
This commit is contained in:
Wentao Ye
2026-04-29 08:30:33 -07:00
committed by GitHub
co-authored by Nick Hill
parent 39a7f4f4e2
commit 51fda1ba44
4 changed files with 30 additions and 22 deletions
+3 -10
View File
@@ -46,7 +46,7 @@ from vllm.tasks import SupportedTask
from vllm.tracing import instrument
from vllm.utils.mem_constants import GiB_bytes
from vllm.utils.mem_utils import MemorySnapshot, format_gib, memory_profiling
from vllm.utils.torch_utils import is_quantized_kv_cache, set_random_seed
from vllm.utils.torch_utils import set_random_seed
from vllm.v1.core.sched.output import GrammarOutput, SchedulerOutput
from vllm.v1.kv_cache_interface import KVCacheConfig, KVCacheSpec
from vllm.v1.outputs import (
@@ -192,15 +192,8 @@ class Worker(WorkerBase):
buffer.data.copy_(self._sleep_saved_buffers[name].data)
self._sleep_saved_buffers = {}
# If the KV cache has just been woken up,
# the internal state of cache_engine must be reset,
# especially the FP8 scaling factor.
if (
(tags is None or "kv_cache" in tags)
and is_quantized_kv_cache(self.cache_config.cache_dtype)
and hasattr(self.model_runner, "init_fp8_kv_scales")
):
self.model_runner.init_fp8_kv_scales()
if tags is None or "kv_cache" in tags:
self.model_runner.post_kv_cache_wake_up()
def _maybe_get_memory_pool_context(self, tag: str) -> AbstractContextManager:
if not self.vllm_config.model_config.enable_sleep_mode: