[Model]Support Step-3.7-Flash (#43859)

Signed-off-by: luotingdan <luotingdan@stepfun.com>
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
Co-authored-by: luotingdan <luotingdan@stepfun.com>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Yu Huang <yuhuang@nvidia.com>
Co-authored-by: Jee Jee Li <jeejeelee@inferact.ai>
This commit is contained in:
ltd0924
2026-05-28 17:01:48 -07:00
committed by GitHub
co-authored by luotingdan Isotr0py Yu Huang Jee Jee Li
parent 325a1ec4fb
commit b690b2bb67
15 changed files with 698 additions and 27 deletions
+5 -1
View File
@@ -38,7 +38,7 @@ logger = init_logger(__name__)
# temporary workaround and better long term solutions are:
# - Add model type to MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS in transformers (better)
# - Fix tokenizer_class on the hub for the affected models (best)
_MODEL_TYPES_WITH_INCORRECT_TOKENIZER_CLASS: set[str] = {"step3_vl"}
_MODEL_TYPES_WITH_INCORRECT_TOKENIZER_CLASS: set[str] = {"step3_vl", "step3p7"}
_VLLM_TOKENIZERS = {
"deepseek_v32": ("deepseek_v32", "DeepseekV32Tokenizer"),
@@ -249,6 +249,10 @@ def get_tokenizer(
tokenizer_cls_ = tokenizer_cls
tokenizer = tokenizer_cls_.from_pretrained(tokenizer_name, *args, **kwargs)
if model_type in _MODEL_TYPES_WITH_INCORRECT_TOKENIZER_CLASS:
from vllm.tokenizers.hf import get_cached_tokenizer
tokenizer = get_cached_tokenizer(tokenizer)
if not tokenizer.is_fast:
logger.warning(
"Using a slow tokenizer. This might cause a significant "