[Bugfix][V1] Read ModelState max_model_len from model config (#58149)

Signed-off-by: Chenglun Hu <chenglunhu@gmail.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
hcl
2026-09-22 19:45:04 +00:00
committed by GitHub
co-authored by mgoin
parent d1f63f2af9
commit 71a2392845
2 changed files with 6 additions and 2 deletions
+5 -1
View File
@@ -59,7 +59,6 @@ class ModelState(ABC):
self.model = model
self.device = device
self.max_model_len = self.model_config.max_model_len
self.max_num_reqs = self.scheduler_config.max_num_seqs
self.max_num_tokens = self.scheduler_config.max_num_batched_tokens
self.inputs_embeds_size = self.model_config.get_inputs_embeds_size()
@@ -99,6 +98,11 @@ class ModelState(ABC):
),
)
@property
def max_model_len(self) -> int:
# Auto-fit can reduce the limit after model state initialization.
return self.model_config.max_model_len
def get_supported_generation_tasks(self) -> tuple[GenerationTask, ...]:
from vllm.model_executor.models.interfaces import (
supports_realtime,