[Bugfix] Skip gs:// model paths for speculator detection (#27846)

Signed-off-by: Peter Schuurman <psch@google.com>
This commit is contained in:
pwschuurman
2025-11-03 14:31:03 +00:00
committed by GitHub
parent 294c805f1d
commit f7d2946e99
3 changed files with 39 additions and 5 deletions
+5 -5
View File
@@ -86,7 +86,7 @@ from vllm.transformers_utils.config import (
is_interleaved,
maybe_override_with_speculators,
)
from vllm.transformers_utils.utils import check_gguf_file, is_s3
from vllm.transformers_utils.utils import check_gguf_file, is_cloud_storage
from vllm.utils.argparse_utils import FlexibleArgumentParser
from vllm.utils.mem_constants import GiB_bytes
from vllm.utils.network_utils import get_ip
@@ -1310,10 +1310,10 @@ class EngineArgs:
# Check if the model is a speculator and override model/tokenizer/config
# BEFORE creating ModelConfig, so the config is created with the target model
# Skip speculator detection for S3 models since HuggingFace cannot load
# configs directly from S3 URLs. S3 models can still use speculators with
# explicit --speculative-config.
if not is_s3(self.model):
# Skip speculator detection for cloud storage models (eg: S3, GCS) since
# HuggingFace cannot load configs directly from S3 URLs. S3 models can still
# use speculators with explicit --speculative-config.
if not is_cloud_storage(self.model):
(self.model, self.tokenizer, self.speculative_config) = (
maybe_override_with_speculators(
model=self.model,