Bump transfomers to 4.54.1 to fix Gemma cache issue. (#8541)

lifuhuang · web-flow · commit 67e53b16f5f1 · 2025-07-30T19:50:54.000-07:00
diff --git a/python/pyproject.toml b/python/pyproject.toml
@@ -45,7 +45,7 @@ runtime_common = [
     "soundfile==0.13.1",
     "scipy",
     "torchao==0.9.0",
-    "transformers==4.54.0",
+    "transformers==4.54.1",
     "timm==1.0.16",
     "uvicorn",
     "uvloop",
diff --git a/test/srt/models/test_generation_models.py b/test/srt/models/test_generation_models.py
@@ -51,8 +51,7 @@ class ModelCase:
 # Popular models that run on the CI
 CI_MODELS = [
     ModelCase("meta-llama/Llama-3.1-8B-Instruct"),
-    # TODO: Gemma is broken by the bug introduced in the latest transformers version, we should restore once its fixed: https://github.com/huggingface/transformers/issues/39711
-    # ModelCase("google/gemma-2-2b"),
+    ModelCase("google/gemma-2-2b"),
 ]
 
 # the complete set of models to test sglang's generation model

Original file line number	Diff line number	Diff line change
`@@ -51,8 +51,7 @@ class ModelCase:`
`51`	`51`	`# Popular models that run on the CI`
`52`	`52`	`CI_MODELS = [`
`53`	`53`	`ModelCase("meta-llama/Llama-3.1-8B-Instruct"),`
`54`		`- # TODO: Gemma is broken by the bug introduced in the latest transformers version, we should restore once its fixed: https://github.com/huggingface/transformers/issues/39711`
`55`		`- # ModelCase("google/gemma-2-2b"),`
	`54`	`+ ModelCase("google/gemma-2-2b"),`
`56`	`55`	`]`
`57`	`56`
`58`	`57`	`# the complete set of models to test sglang's generation model`