[CI] Bump Transformers version to 5.21.1

requirements/common.txt

@@ -8,7 +8,7 @@ certifi >= 2026.10.19
 tqdm
 blake3
 py-cpuinfo
-transformers >= 5.16.1, < 5.21.0
+transformers >= 5.16.1, < 5.22.0
 huggingface_hub >= 1.31.0  # Required for the `huggingface_hub.utils.httpx` re-export
 tokenizers >= 0.23.2  # Required for fast incremental detokenization.
 safetensors >= 0.6.2  # MXFP4/MXFP6 dtype support (F8_E8M0, F4) added in 0.6.0: https://github.com/huggingface/safetensors/pull/611

requirements/test/cpu.txt

@@ -1227,7 +1227,7 @@ traitlets==5.16.1
     # via
     #   jupyter-client
     #   jupyter-core
-transformers==5.20.2
+transformers==5.21.1
     # via
     #   -r requirements/test/../common.txt
     #   -r requirements/test/cuda.in

requirements/test/cuda.in

@@ -40,7 +40,7 @@ datamodel_code_generator>=0.81.0 # required for minicpm3 test
 nltk>=3.10.3 # pin transitive rouge-score/lm-eval dep for known GHSAs
 lm-eval[api]>=0.4.12 # required for model evaluation test
 mteb[bm25s]>=2, <3 # required for mteb test
-transformers==5.20.2
+transformers==5.21.1
 schemathesis>=4.0.0 # Required for openai schema test.
 # quantization
 buildkite-test-collector==0.1.9

requirements/test/cuda.txt

@@ -1337,7 +1337,7 @@ traitlets==5.16.1
     # via
     #   jupyter-client
     #   jupyter-core
-transformers==5.20.2
+transformers==5.21.1
     # via
     #   -c requirements/common.txt
     #   -r requirements/test/../common.txt

requirements/test/nightly-torch.txt

@@ -29,7 +29,7 @@ opencv-python-headless >= 4.13.0 # required for video test
 datamodel_code_generator>=0.81.0 # required for minicpm3 test
 lm-eval[api]>=0.4.12 # required for model evaluation test
 mteb[bm25s]>=2, <3 # required for mteb test
-transformers==5.20.2
+transformers==5.21.1
 schemathesis>=4.0.0 # Required for openai schema test.
 # quantization
 buildkite-test-collector==0.1.9

requirements/test/rocm.in

@@ -36,7 +36,7 @@ datamodel_code_generator>=0.81.0 # required for minicpm3 test
 nltk>=3.10.3 # pin transitive rouge-score/lm-eval dep for known GHSAs
 lm-eval[api]>=0.4.12 # required for model evaluation test
 mteb[bm25s]>=2, <3 # required for mteb test
-transformers==5.20.2
+transformers==5.21.1
 schemathesis>=4.0.0 # Required for openai schema test
 # quantization
 bitsandbytes==0.50.0

requirements/test/rocm.txt

@@ -1287,7 +1287,7 @@ traitlets==5.16.1
     # via
     #   jupyter-client
     #   jupyter-core
-transformers==5.20.2
+transformers==5.21.1
     # via
     #   -c requirements/common.txt
     #   -r requirements/test/../common.txt

requirements/test/xpu.in

@@ -21,7 +21,7 @@ nltk>=3.10.3 # pin transitive rouge-score/lm-eval dep for known GHSAs
 modelscope<1.38
 runai-model-streamer[s3,gcs,azure]>=0.15.7
 tensorizer==2.10.1
-transformers==5.20.2
+transformers==5.21.1
 
 # --- Audio Processing ---
 av

requirements/test/xpu.txt

@@ -1156,7 +1156,7 @@ traitlets==5.16.1
     # via
     #   jupyter-client
     #   jupyter-core
-transformers==5.20.2
+transformers==5.21.1
     # via
     #   -c requirements/common.txt
     #   -r requirements/test/../common.txt

vllm/model_executor/models/minicpmv4_6.py

@@ -7,6 +7,8 @@
 
 import numpy as np
 import torch
+import transformers
+from packaging.version import Version
 from PIL import Image as PILImage
 from torch import nn
 from transformers import MiniCPMV4_6Config
@@ -122,6 +124,12 @@ def _expand_flags(value: object, num_items: int) -> list[bool | None]:
     return [any(bool(v) for v in raw)] * num_items
 
 
+# Transformers >= 5.21 sizes images by the `downsample_mode` merge factor
+RESIZE_TAKES_DOWNSAMPLE_MODE = Version(transformers.__version__) >= Version(
+    "5.21.0.dev0"
+)
+
+
 def _vision_downsample_mode(flag: bool | None) -> str | None:
     if flag is None:
         return None
@@ -739,6 +747,9 @@ def _compute_visual_tokens(
         scale_res = image_processor.scale_resolution
         downsample_mode = self._get_downsample_mode(downsample_mode)
         token_divisor = 4 if downsample_mode == "4x" else 16
+        resize_kwargs = (
+            {"downsample_mode": downsample_mode} if RESIZE_TAKES_DOWNSAMPLE_MODE else {}
+        )
 
         # vLLM ImageSize is (width, height); transformers expects (height, width)
         hf_image_size = (image_size.height, image_size.width)
@@ -762,6 +773,7 @@ def _compute_visual_tokens(
                 scale_res,
                 patch_size,
                 allow_upscale=True,
+                **resize_kwargs,
             )
             source_tokens = (
                 best_size[0] * best_size[1] // (patch_size * patch_size * token_divisor)
@@ -772,6 +784,7 @@ def _compute_visual_tokens(
             hf_image_size,
             scale_res,
             patch_size,
+            **resize_kwargs,
         )
         source_tokens = (
             best_resize[0] * best_resize[1] // (patch_size * patch_size * token_divisor)
@@ -782,6 +795,7 @@ def _compute_visual_tokens(
             scale_res,
             patch_size,
             allow_upscale=True,
+            **resize_kwargs,
         )
         patch_w = refine_size[0] // grids[0]
         patch_h = refine_size[1] // grids[1]