[Misc] Remove unnecessary Transformers version guards (#57556)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-09-18 15:01:35 +00:00
committed by GitHub
parent 1cdf1689e5
commit dbf4b89cf2
11 changed files with 15 additions and 146 deletions
@@ -3,8 +3,6 @@
import pytest
import torch
from packaging.version import Version
from transformers import __version__ as TRANSFORMERS_VERSION
from vllm.platforms import current_platform
@@ -162,16 +160,6 @@ def test_models(
if prompt_embeds is not None:
embed = hf_model.model.get_input_embeddings()(token_ids)
if "gemma" in model.lower() and (
Version(TRANSFORMERS_VERSION) < Version("5.3.0.dev0")
):
# For Gemma 1/2 models with Transformers 5.4.0+, the prompt
# embeddings are normalised in `get_prompt_embeddings`,
# like Gemma 3. For older versions, we need to manually normalise.
embed_scale = hf_model.config.hidden_size**0.5
normalizer = torch.tensor(embed_scale, dtype=embed.dtype)
embed *= normalizer
# MiniCPM models apply scale_emb to embeddings internally.
# vLLM expects pre-scaled embeddings when using inputs_embeds.
if model in EMBED_SCALING_MODELS: