[Bugfix] Honor OPT token embedding width for prompt embeddings (#61736)

Signed-off-by: maithilijoshi20 <maithilij2003@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
maithilijoshi20
2026-10-16 04:41:50 +00:00
committed by GitHub
co-authored by Cyrus Leung
parent b466fbacd2
commit 92dc3dd393
9 changed files with 60 additions and 9 deletions
+8 -4
View File
@@ -15,6 +15,7 @@ from vllm.tokenizers.detokenizer_utils import (
)
from vllm.tokenizers.mistral import MistralTokenizer
from vllm.tokenizers.protocol import TokenizerLike
from vllm.transformers_utils.repo_utils import with_retry
from vllm.v1.engine import EngineCoreRequest
from vllm.v1.engine.detokenizer import (
FastIncrementalDetokenizer,
@@ -95,10 +96,13 @@ def _run_incremental_decode(
@pytest.fixture
def tokenizer(tokenizer_name):
return (
MistralTokenizer.from_pretrained(tokenizer_name)
if "mistral" in tokenizer_name
else AutoTokenizer.from_pretrained(tokenizer_name)
return with_retry(
lambda: (
MistralTokenizer.from_pretrained(tokenizer_name)
if "mistral" in tokenizer_name
else AutoTokenizer.from_pretrained(tokenizer_name)
),
log_msg=f"Error loading tokenizer {tokenizer_name}",
)