huggingface · simondanielsson · Sep 2, 2025 · Sep 2, 2025 · Sep 2, 2025 · Sep 2, 2025
diff --git a/docs/source/onnx/overview.mdx b/docs/source/onnx/overview.mdx
@@ -52,6 +52,8 @@ Supported architectures from [🤗 Transformers](https://huggingface.co/docs/tra
 - ESM
 - Falcon
 - Flaubert
+- Gemma
+- Gemma3
 - GPT-2
 - GPT-BigCode
 - GPT-J

diff --git a/optimum/exporters/onnx/model_configs.py b/optimum/exporters/onnx/model_configs.py
@@ -507,6 +507,17 @@ class GemmaOnnxConfig(LlamaOnnxConfig):
     MIN_TRANSFORMERS_VERSION = version.parse("4.38.0")
 
 
+@register_tasks_manager_onnx("gemma3", *[*COMMON_TEXT_GENERATION_TASKS, "text-classification"])
+class Gemma3OnnxConfig(LlamaOnnxConfig):
+    DUMMY_INPUT_GENERATOR_CLASSES = (
+        DummyTextInputGenerator,
+        DummyVisionInputGenerator,
+    )
+    DUMMY_PKV_GENERATOR_CLASS = GemmaDummyPastKeyValuesGenerator
+    NORMALIZED_CONFIG_CLASS = NormalizedConfigManager.get_normalized_config_class("gemma3")
+    MIN_TRANSFORMERS_VERSION = version.parse("4.52.0.dev0")
+
+
 @register_tasks_manager_onnx("nemotron", *COMMON_TEXT_GENERATION_TASKS)
 class NemotronOnnxConfig(GemmaOnnxConfig):
     MIN_TRANSFORMERS_VERSION = version.parse("4.48.0")  # More stable version than 4.44.0

diff --git a/optimum/exporters/onnx/utils.py b/optimum/exporters/onnx/utils.py
@@ -75,6 +75,7 @@
     "cohere",
     "falcon",
     "gemma",
+    "gemma3",
     "gpt2",
     "gpt_bigcode",
     "gpt_neo",

diff --git a/tests/exporters/onnx/utils_tests.py b/tests/exporters/onnx/utils_tests.py
@@ -103,6 +103,7 @@
     },
     "flaubert": "hf-internal-testing/tiny-random-flaubert",
     "gemma": "fxmarty/tiny-random-GemmaForCausalLM",
+    "gemma3": "hf-internal-testing/tiny-random-Gemma3ForConditionalGeneration",
     "glpn": "hf-internal-testing/tiny-random-GLPNModel",
     "gpt2": "hf-internal-testing/tiny-random-gpt2",
     "gpt_bigcode": "hf-internal-testing/tiny-random-GPTBigCodeModel",

diff --git a/tests/onnxruntime/test_decoder.py b/tests/onnxruntime/test_decoder.py
@@ -32,6 +32,7 @@
     BloomOnnxConfig,
     CohereOnnxConfig,
     DeepSeekV3OnnxConfig,
+    Gemma3OnnxConfig,
     GemmaOnnxConfig,
     GraniteOnnxConfig,
     HeliumOnnxConfig,
@@ -108,6 +109,8 @@ class ORTModelForCausalLMIntegrationTest(ORTModelTestMixin):
         SUPPORTED_ARCHITECTURES.append("qwen2")
     if is_transformers_version(">=", str(GemmaOnnxConfig.MIN_TRANSFORMERS_VERSION)):
         SUPPORTED_ARCHITECTURES.append("gemma")
+    if is_transformers_version(">=", str(Gemma3OnnxConfig.MIN_TRANSFORMERS_VERSION)):
+        SUPPORTED_ARCHITECTURES.append("gemma3")
     if is_transformers_version(">=", str(MPTOnnxConfig.MIN_TRANSFORMERS_VERSION)):
         SUPPORTED_ARCHITECTURES.append("mpt")
     if is_transformers_version(">=", str(NemotronOnnxConfig.MIN_TRANSFORMERS_VERSION)):

diff --git a/tests/onnxruntime/testing_utils.py b/tests/onnxruntime/testing_utils.py
@@ -68,6 +68,7 @@
     "flaubert": "hf-internal-testing/tiny-random-flaubert",
     "flux": "optimum-internal-testing/tiny-random-flux",
     "gemma": "fxmarty/tiny-random-GemmaForCausalLM",
+    "gemma3": "hf-internal-testing/tiny-random-Gemma3ForConditionalGeneration",
     "gpt2": "hf-internal-testing/tiny-random-GPT2LMHeadModel",
     "gpt_bigcode": "hf-internal-testing/tiny-random-GPTBigCodeModel",
     "gpt_bigcode-multi_query-False": "optimum-internal-testing/tiny-random-gpt_bigcode-multi_query-False",
-Original file line number
+Diff line change
@@ Expand Up @@
     - ESM
     - Falcon
     - Flaubert
+    - Gemma
+    - Gemma3
     - GPT-2
     - GPT-BigCode
     - GPT-J
@@ Expand Down @@