[Model] Add step3 vl

docs/models/supported_models.md

@@ -598,6 +598,7 @@ Specified using `--task generate`.
 | `Qwen2_5OmniThinkerForConditionalGeneration` | Qwen2.5-Omni                                                             | T + I<sup>E+</sup> + V<sup>E+</sup> + A<sup>+</sup>                   | `Qwen/Qwen2.5-Omni-7B`                                                                                                                                  |                        | ✅︎                          | ✅︎\*                  |
 | `SkyworkR1VChatModel`                        | Skywork-R1V-38B                                                          | T + I                                                                 | `Skywork/Skywork-R1V-38B`                                                                                                                               |                        | ✅︎                          | ✅︎                    |
 | `SmolVLMForConditionalGeneration`            | SmolVLM2                                                                 | T + I                                                                 | `SmolVLM2-2.2B-Instruct`                                                                                                                                | ✅︎                     |                             | ✅︎                    |
+| `Step3VLForConditionalGeneration` | Step3-VL | T + I<sup>+</sup> | `stepfun-ai/step3` | | ✅︎ | ✅︎ |
 | `TarsierForConditionalGeneration`            | Tarsier                                                                  | T + I<sup>E+</sup>                                                    | `omni-search/Tarsier-7b`,`omni-search/Tarsier-34b`                                                                                                      |                        | ✅︎                          | ✅︎                    |
 | `Tarsier2ForConditionalGeneration`<sup>^</sup> | Tarsier2                                                                 | T + I<sup>E+</sup> + V<sup>E+</sup>                                                | `omni-research/Tarsier2-Recap-7b`,`omni-research/Tarsier2-7b-0115`                                                                                      |                        | ✅︎                          | ✅︎                    |
 

tests/models/registry.py

@@ -266,6 +266,9 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
     "StableLMEpochForCausalLM": _HfExamplesInfo(os.path.join(models_path_prefix,"stabilityai/stablelm-zephyr-3b")),  # noqa: E501
     "StableLmForCausalLM": _HfExamplesInfo(os.path.join(models_path_prefix,"stabilityai/stablelm-3b-4e1t")),
     "Starcoder2ForCausalLM": _HfExamplesInfo(os.path.join(models_path_prefix,"bigcode/starcoder2-3b")),
+    "Step3TextForCausalLM": _HfExamplesInfo(os.path.join(models_path_prefix, "stepfun-ai/step3"),
+                                            trust_remote_code=True,
+                                            is_available_online=False),
     "SolarForCausalLM": _HfExamplesInfo(os.path.join(models_path_prefix,"upstage/solar-pro-preview-instruct")),
     "TeleChat2ForCausalLM": _HfExamplesInfo(os.path.join(models_path_prefix,"Tele-AI/TeleChat2-3B"),
                                             trust_remote_code=True),
@@ -423,6 +426,9 @@ _MULTIMODAL_EXAMPLE_MODELS = {
     "Qwen2_5OmniForConditionalGeneration": _HfExamplesInfo(os.path.join(models_path_prefix, "Qwen/Qwen2.5-Omni-7B-AWQ")),  # noqa: E501
     "SkyworkR1VChatModel": _HfExamplesInfo(os.path.join(models_path_prefix, "Skywork/Skywork-R1V-38B")),
     "SmolVLMForConditionalGeneration": _HfExamplesInfo(os.path.join(models_path_prefix, "HuggingFaceTB/SmolVLM2-2.2B-Instruct")),  # noqa: E501
+    "Step3VLForConditionalGeneration": _HfExamplesInfo(os.path.join(models_path_prefix, "stepfun-ai/step3"),
+                                                        trust_remote_code=True,
+                                                        is_available_online=False),
     "UltravoxModel": _HfExamplesInfo(os.path.join(models_path_prefix, "fixie-ai/ultravox-v0_5-llama-3_2-1b"),  # noqa: E501
                                      trust_remote_code=True),
     "TarsierForConditionalGeneration": _HfExamplesInfo("omni-research/Tarsier-7b",  # noqa: E501

vllm/entrypoints/openai/tool_parsers/__init__.py

@@ -15,6 +15,7 @@ from .minimax_tool_parser import MinimaxToolParser
 from .mistral_tool_parser import MistralToolParser
 from .phi4mini_tool_parser import Phi4MiniJsonToolParser
 from .pythonic_tool_parser import PythonicToolParser
+from .step3_tool_parser import Step3ToolParser
 from .xlam_tool_parser import xLAMToolParser
 
 __all__ = [
@@ -31,6 +32,7 @@ __all__ = [
     "PythonicToolParser", 
     "Phi4MiniJsonToolParser",
     "DeepSeekV3ToolParser", 
+    "Step3ToolParser",
     "xLAMToolParser", 
     "MinimaxToolParser",
     "Glm4MoeModelToolParser",

vllm/model_executor/models/registry.py

@@ -120,6 +120,7 @@ _TEXT_GENERATION_MODELS = {
     "Qwen3ForCausalLM": ("qwen3", "Qwen3ForCausalLM"),
     "Qwen3MoeForCausalLM": ("qwen3_moe", "Qwen3MoeForCausalLM"),
     "RWForCausalLM": ("falcon", "FalconForCausalLM"),
+    "Step3TextForCausalLM": ("step3_text", "Step3TextForCausalLM"),
     "StableLMEpochForCausalLM": ("stablelm", "StablelmForCausalLM"),
     "StableLmForCausalLM": ("stablelm", "StablelmForCausalLM"),
     "Starcoder2ForCausalLM": ("starcoder2", "Starcoder2ForCausalLM"),
@@ -228,6 +229,7 @@ _MULTIMODAL_MODELS = {
     "Qwen2_5OmniModel": ("qwen2_5_omni_thinker", "Qwen2_5OmniThinkerForConditionalGeneration"),  # noqa: E501
     "Qwen2_5OmniForConditionalGeneration": ("qwen2_5_omni_thinker", "Qwen2_5OmniThinkerForConditionalGeneration"),  # noqa: E501
     "UltravoxModel": ("ultravox", "UltravoxModel"),
+    "Step3VLForConditionalGeneration": ("step3_vl", "Step3VLForConditionalGeneration"),  # noqa: E501
     "Phi4MMForCausalLM": ("phi4mm", "Phi4MMForCausalLM"),
     "TarsierForConditionalGeneration": ("tarsier", "TarsierForConditionalGeneration"),  # noqa: E501
     "Tarsier2ForConditionalGeneration": ("qwen2_vl", "Tarsier2ForConditionalGeneration"),  # noqa: E501

vllm/reasoning/__init__.py

@@ -6,6 +6,7 @@ from .deepseek_r1_reasoning_parser import DeepSeekR1ReasoningParser
 from .glm4_moe_reasoning_parser import Glm4MoeModelReasoningParser
 from .granite_reasoning_parser import GraniteReasoningParser
 from .qwen3_reasoning_parser import Qwen3ReasoningParser
+from .step3_reasoning_parser import Step3ReasoningParser
 
 __all__ = [
     "ReasoningParser",
@@ -14,4 +15,5 @@ __all__ = [
     "GraniteReasoningParser",
     "Qwen3ReasoningParser",
     "Glm4MoeModelReasoningParser",
+    "Step3ReasoningParser",
 ]

vllm/transformers_utils/config.py

@@ -39,6 +39,7 @@ from vllm.transformers_utils.configs import (ChatGLMConfig, Cohere2Config,
                                              MLPSpeculatorConfig, MPTConfig,
                                              NemotronConfig, NVLM_D_Config,
                                              OvisConfig, RWConfig,
+                                             Step3TextConfig, Step3VLConfig, 
                                              SkyworkR1VChatConfig, SolarConfig,
                                              Telechat2Config, UltravoxConfig)
 # yapf: enable
@@ -97,6 +98,8 @@ _CONFIG_REGISTRY: dict[str, type[PretrainedConfig]] = {
     "skywork_chat": SkyworkR1VChatConfig,
     "telechat": Telechat2Config,
     "ultravox": UltravoxConfig,
+    "step3_vl": Step3VLConfig,
+    "step3_text": Step3TextConfig,
     **_CONFIG_REGISTRY_OVERRIDE_HF
 }
 

vllm/transformers_utils/configs/__init__.py

@@ -28,6 +28,9 @@ from vllm.transformers_utils.configs.ovis import OvisConfig
 from vllm.transformers_utils.configs.skyworkr1v import SkyworkR1VChatConfig
 from vllm.transformers_utils.configs.solar import SolarConfig
 from vllm.transformers_utils.configs.telechat2 import Telechat2Config
+from vllm.transformers_utils.configs.step3_vl import (Step3TextConfig,
+                                                      Step3VisionEncoderConfig,
+                                                      Step3VLConfig)
 from vllm.transformers_utils.configs.ultravox import UltravoxConfig
 
 __all__ = [
@@ -56,4 +59,7 @@ __all__ = [
     "SolarConfig",
     "Telechat2Config",
     "UltravoxConfig",
+    "Step3VLConfig",
+    "Step3VisionEncoderConfig",
+    "Step3TextConfig",
 ]