Support vLLM 0.28 (issue #84)

.github/workflows/gpu-tests.yaml

@@ -106,17 +106,17 @@ jobs:
     # this is evaluated before any step runs.
     name: GPU Tests (${{ matrix.label }} · ${{ inputs.suite || 'full' }})
     # granite-switch supports two mutually-exclusive vLLM lines and both must be
-    # tested. They cannot share a venv (pyproject declares vllm19/vllm20 as
+    # tested. They cannot share a venv (pyproject declares vllm19/vllm28 as
     # conflicting groups), so each gets its own cluster pod. Only the dev* groups
-    # used because the bare vllm19/vllm20 groups omit pytest.
+    # used because the bare vllm19/vllm28 groups omit pytest.
     strategy:
-      fail-fast: false   # a vllm19 failure must not hide the vllm20 result
+      fail-fast: false   # a vllm19 failure must not hide the vllm28 result
       matrix:
         include:
           - label: vllm19
             dep_group: dev
-          - label: vllm20
-            dep_group: dev-vllm20
+          - label: vllm28
+            dep_group: dev-vllm28
     runs-on: [self-hosted, gpu]
 
     # MUST be set explicitly. GitHub's default is 360 minutes, which is exactly the

README.md

@@ -47,15 +47,15 @@ Other install options depending on your use case:
 ```bash
 pip install "granite-switch[compose]"   # Compose modular models
 pip install "granite-switch[hf]"        # HuggingFace inference
-pip install "granite-switch[vllm20]"    # vLLM 0.20+ (requires CUDA 13+)
+pip install "granite-switch[vllm28]"    # vLLM 0.28+ (requires CUDA 13+)
 pip install "granite-switch[dev]"       # Everything
 ```
 
 Requires Python 3.10+ and PyTorch 2.0+.
 
-> **vLLM version note:** This project currently defaults to vLLM 0.19.1 due to vLLM 0.20's
-> dependency on CUDA 13.0+ (via PyTorch 2.11), which is incompatible with many existing
-> environments running CUDA 12.x drivers. Use `.[vllm20]` if your environment supports CUDA 13+.
+> **vLLM version note:** This project currently defaults to vLLM 0.19.1 due to vLLM 0.28's
+> dependency on CUDA 13.0+ (via PyTorch 2.13), which is incompatible with many existing
+> environments running CUDA 12.x drivers. Use `.[vllm28]` if your environment supports CUDA 13+.
 
 ### Compose a Model
 

docs/AUDIO.md

@@ -19,11 +19,11 @@ non-16 kHz input:
 
 ```bash
 # Serving an audio-enabled checkpoint
-uv sync --extra vllm --extra audio     # or --extra vllm20 --extra audio
+uv sync --extra vllm --extra audio     # or --extra vllm28 --extra audio
 
 # Development / running the test suite (the dev groups include audio already)
 uv sync --group dev                    # vLLM 0.19.x
-uv sync --group dev-vllm20             # vLLM 0.20.x
+uv sync --group dev-vllm28             # vLLM 0.28.x
 ```
 
 ## Building an audio-enabled checkpoint

pyproject.toml

@@ -11,7 +11,7 @@ license = "Apache-2.0"
 requires-python = ">=3.11,<3.14"
 dependencies = [
     "torch>=2.10.0",
-    "transformers>=5.5.1,<5.10.0",
+    "transformers>=5.5.1,<5.16.0",
 ]
 
 [project.urls]
@@ -21,7 +21,7 @@ Documentation = "https://github.com/generative-computing/granite-switch/tree/mai
 [project.optional-dependencies]
 hf = ["accelerate>=0.20.0"]
 vllm = ["vllm>=0.19.1,<0.20.0"]
-vllm20 = ["vllm>=0.20.0,<0.21.0"]
+vllm28 = ["vllm>=0.28.0,<0.29.0"]
 compose = ["huggingface_hub", "pyyaml", "tqdm", "safetensors"]
 build = ["huggingface_hub", "pyyaml", "tqdm", "safetensors"]  # Backward compatibility alias for compose
 # Audio (ASR) decode + resample. Reuse vLLM's own audio deps (unversioned, so it
@@ -57,32 +57,32 @@ markers = [
 
 [dependency-groups]
 vllm19     = ["vllm>=0.19.1,<0.20.0"]
-vllm20     = ["vllm>=0.20.0,<0.21.0"]
+vllm28     = ["vllm>=0.28.0,<0.29.0"]
 # `audio` is included so the audio tests can actually run: the ASR path needs
 # vLLM's audio deps (av/soundfile/resampy) at runtime, and no group pulled them
 # in before (integration tests failed with ModuleNotFoundError on a synced pod).
 dev        = ["pytest", "pytest-cov", { include-group = "vllm19" }, "granite-switch[hf,compose,audio]"]
-dev-vllm20 = ["pytest", "pytest-cov", { include-group = "vllm20" }, "granite-switch[hf,compose,audio]"]
+dev-vllm28 = ["pytest", "pytest-cov", { include-group = "vllm28" }, "granite-switch[hf,compose,audio]"]
 test       = ["pytest", "pytest-cov", "bitsandbytes", "optimum-quanto", { include-group = "dev" }]
 
 [tool.uv]
 default-groups = ["vllm19"]
 conflicts = [
     # group-vs-group
-    [{ group = "vllm19" },     { group = "vllm20" }],
-    [{ group = "dev" },        { group = "vllm20" }],
-    [{ group = "dev" },        { group = "dev-vllm20" }],
-    [{ group = "dev-vllm20" }, { group = "vllm19" }],
+    [{ group = "vllm19" },     { group = "vllm28" }],
+    [{ group = "dev" },        { group = "vllm28" }],
+    [{ group = "dev" },        { group = "dev-vllm28" }],
+    [{ group = "dev-vllm28" }, { group = "vllm19" }],
     # group-vs-extra
-    [{ group = "vllm19" },     { extra = "vllm20" }],
-    [{ group = "vllm20" },     { extra = "vllm" }],
-    [{ group = "vllm20" },     { extra = "tutorials" }],
-    [{ group = "dev" },        { extra = "vllm20" }],
-    [{ group = "dev-vllm20" }, { extra = "vllm" }],
-    [{ group = "dev-vllm20" }, { extra = "tutorials" }],
+    [{ group = "vllm19" },     { extra = "vllm28" }],
+    [{ group = "vllm28" },     { extra = "vllm" }],
+    [{ group = "vllm28" },     { extra = "tutorials" }],
+    [{ group = "dev" },        { extra = "vllm28" }],
+    [{ group = "dev-vllm28" }, { extra = "vllm" }],
+    [{ group = "dev-vllm28" }, { extra = "tutorials" }],
     # extra-vs-extra
-    [{ extra = "vllm" },       { extra = "vllm20" }],
-    [{ extra = "tutorials" },  { extra = "vllm20" }],
+    [{ extra = "vllm" },       { extra = "vllm28" }],
+    [{ extra = "tutorials" },  { extra = "vllm28" }],
 ]
 
 [tool.coverage.run]

tests/integration/test_audio_uncached_processor.py

@@ -55,7 +55,7 @@ def _cache_disabling_kwargs():
     """LLM kwargs that turn off the multimodal processor cache.
 
     The knob was renamed across the vLLM range this project supports (0.19.x and
-    0.20.x are both allowed in pyproject): older builds expose
+    0.28.x are both allowed in pyproject): older builds expose
     ``disable_mm_preprocessor_cache``, newer ones ``mm_processor_cache_gb``.
     Returns an empty dict when neither exists, so the caller can skip rather than
     silently exercise the cached path.

tutorials/PREREQUISITES.md

@@ -42,7 +42,7 @@ For production inference with vLLM:
 
 ```bash
 pip install "granite-switch[vllm]"    # CUDA 12.x
-pip install "granite-switch[vllm20]"  # CUDA 13+ (requires PyTorch 2.11+)
+pip install "granite-switch[vllm28]"  # CUDA 13+ (requires PyTorch 2.13+)
 ```
 
 This includes: