Add GLM Export (#51)

Add ONNX export for GLM

Author: Abdennacer-Badaoui

docs/source/onnx/overview.mdx

@@ -52,6 +52,7 @@ Supported architectures from [🤗 Transformers](https://huggingface.co/docs/tra
 - ESM
 - Falcon
 - Flaubert
+- GLM
 - GPT-2
 - GPT-BigCode
 - GPT-J

optimum/exporters/onnx/model_configs.py

@@ -456,6 +456,11 @@ class CohereOnnxConfig(LlamaOnnxConfig):
     _MODEL_PATCHER = CohereModelPatcher
 
 
+@register_tasks_manager_onnx("glm", *COMMON_TEXT_GENERATION_TASKS)
+class GLMOnnxConfig(LlamaOnnxConfig):
+    MIN_TRANSFORMERS_VERSION = version.parse("4.46.0")
+
+
 @register_tasks_manager_onnx("helium", *COMMON_TEXT_GENERATION_TASKS)
 class HeliumOnnxConfig(LlamaOnnxConfig):
     MIN_TRANSFORMERS_VERSION = version.parse("4.49.0")

optimum/exporters/onnx/utils.py

@@ -75,12 +75,14 @@
     "cohere",
     "falcon",
     "gemma",
+    "glm",
     "gpt2",
     "gpt_bigcode",
     "gpt_neo",
     "gpt_neox",
     "gpt_oss",
     "gptj",
+    "granite",
     "helium",
     "imagegpt",
     "internlm2",
@@ -92,7 +94,6 @@
     "qwen2",
     "qwen3",
     "qwen3_moe",
-    "granite",
     "smollm3",
     "stablelm",
     "olmo2",

optimum/onnxruntime/modeling_decoder.py

@@ -246,6 +246,8 @@ def __init__(
             "deepseek_v3",
             "cohere",
             "gemma",
+            "glm",
+            "granite",
             "gpt_oss",
             "helium",
             "mistral",
@@ -254,7 +256,6 @@ def __init__(
             "qwen2",
             "qwen3",
             "qwen3_moe",
-            "granite",
             "smollm3",
             "stablelm",
         }:

tests/exporters/onnx/utils_tests.py

@@ -103,6 +103,7 @@
     },
     "flaubert": "hf-internal-testing/tiny-random-flaubert",
     "gemma": "fxmarty/tiny-random-GemmaForCausalLM",
+    "glm": "hf-internal-testing/tiny-random-GlmForCausalLM",
     "glpn": "hf-internal-testing/tiny-random-GLPNModel",
     "gpt2": "hf-internal-testing/tiny-random-gpt2",
     "gpt_bigcode": "hf-internal-testing/tiny-random-GPTBigCodeModel",

tests/onnxruntime/test_decoder.py

@@ -33,6 +33,7 @@
     CohereOnnxConfig,
     DeepSeekV3OnnxConfig,
     GemmaOnnxConfig,
+    GLMOnnxConfig,
     GPTOssOnnxConfig,
     GraniteOnnxConfig,
     HeliumOnnxConfig,
@@ -112,6 +113,8 @@ 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(GLMOnnxConfig.MIN_TRANSFORMERS_VERSION)):
+        SUPPORTED_ARCHITECTURES.append("glm")
     if is_transformers_version(">=", str(MPTOnnxConfig.MIN_TRANSFORMERS_VERSION)):
         SUPPORTED_ARCHITECTURES.append("mpt")
     if is_transformers_version(">=", str(NemotronOnnxConfig.MIN_TRANSFORMERS_VERSION)):

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",
+    "glm": "hf-internal-testing/tiny-random-GlmForCausalLM",
     "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",