feat: Support Florence-2 Transformers-native architecture (transformers ≥ 4.56.1 / 5.x)

README.md

@@ -15,6 +15,13 @@ This plugin requires [uv](https://docs.astral.sh/uv/) for package management. If
 ```bash
 curl -LsSf https://astral.sh/uv/install.sh | sh
 ```
+### From Git
+
+Install from git:
+
+```bash
+pip install git+https://github.com/vllm-project/bart-plugin
+```
 
 ### From Source
 
@@ -186,11 +193,14 @@ Notes:
 ```
 bart-plugin/
 ├── vllm_bart_plugin/
-│   ├── __init__.py          # Plugin registration
-│   └── bart.py              # BART model implementation
-├── setup.py                 # Package configuration and entry points
-├── README.md                # This file
-└── LICENSE                  # License file
+│   ├── __init__.py            # Plugin registration
+│   └── bart.py                # BART model implementation
+│   └── florence2.py           # Florence-2 model implementation
+├── setup.py                   # Package configuration and entry points
+├── README.md                  # This file
+└── LICENSE                    # License file
+└── example_bart_usage.py      # Example usage script for BART
+└── example_florence2_usage.py # Example usage script for Florence-2
 ```
 
 ### Running Tests

example_florence2_usage.py

@@ -5,28 +5,23 @@
 This script demonstrates how to use Florence-2 models with vLLM
 after installing the BART plugin.
 """
-import vllm_bart_plugin
+
 from vllm import LLM, SamplingParams
 from vllm.assets.image import ImageAsset
 
 
 def main():
     """Run Florence-2 model examples."""
-    model_name = "microsoft/Florence-2-large"
-    tokenizer_name = "Isotr0py/Florence-2-tokenizer"
+    model_name = "florence-community/Florence-2-large-ft"
 
     llm = LLM(
         model=model_name,
-        tokenizer=tokenizer_name,
         mm_processor_cache_gb=0,
-        trust_remote_code=True,
         enforce_eager=True,
     )
     params = SamplingParams(
         temperature=0.0,
         max_tokens=20,
-        # repetition_penalty is needed to prevent <s> repetition
-        repetition_penalty=1.5,
         # skip_special_tokens=False is needed to present
         # grounding tokens like <loc_0><loc_1>
         skip_special_tokens=False,
@@ -60,4 +55,4 @@ def main():
 
 
 if __name__ == "__main__":
-    main()
+    main()

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "vllm-bart-plugin"
-version = "0.2.0"
+version = "0.3.0"
 description = "BART model plugin for vLLM"
 readme = "README.md"
 requires-python = ">=3.10"
@@ -26,9 +26,9 @@ classifiers = [
 ]
 
 dependencies = [
-    "vllm>=0.14.0",
+    "vllm>=0.16.0",
     "torch>=2.9.0",
-    "transformers>=4.56.0,<5",
+    "transformers>=4.56.0,<6",
 ]
 
 [project.optional-dependencies]
@@ -62,8 +62,20 @@ include = '\.pyi?$'
 profile = "black"
 line_length = 88
 
+[tool.pytest.ini_options]
+markers = [
+    "slow: marks tests requiring a GPU and full model download (deselect with '-m \"not slow\"')",
+]
+
 [tool.mypy]
 python_version = "3.10"
 warn_return_any = true
 warn_unused_configs = true
 ignore_missing_imports = true
+
+[dependency-groups]
+dev = [
+    "black>=26.1.0",
+    "isort>=8.0.1",
+    "pytest>=9.0.2",
+]