Merge tag 'v0.9.1' into v0.9.1-ori

tests/plugins/vllm_add_dummy_model/vllm_add_dummy_model/my_llava.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from typing import Optional
 

tests/plugins/vllm_add_dummy_model/vllm_add_dummy_model/my_opt.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from typing import Optional
 

tests/plugins/vllm_add_dummy_platform/setup.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from setuptools import setup
 

tests/plugins/vllm_add_dummy_platform/vllm_add_dummy_platform/__init__.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from typing import Optional
 

tests/plugins/vllm_add_dummy_platform/vllm_add_dummy_platform/dummy_attention_backend.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from vllm.attention.backends.flash_attn import FlashAttentionBackend
 

tests/plugins/vllm_add_dummy_platform/vllm_add_dummy_platform/dummy_platform.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from vllm.platforms.cuda import CudaPlatform
 

tests/plugins_tests/conftest.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 import pytest
 
 

tests/plugins_tests/test_platform_plugins.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import pytest
 import torch

tests/plugins_tests/test_scheduler_plugins.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import pytest
 

tests/pplx_utils.py

+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import dataclasses
+import os
+import traceback
+from typing import Callable
+
+import torch
+from torch.multiprocessing import (
+    spawn)  # pyright: ignore[reportPrivateImportUsage]
+from typing_extensions import Concatenate, ParamSpec
+
+P = ParamSpec("P")
+
+
+@dataclasses.dataclass
+class ProcessGroupInfo:
+    world_size: int
+    world_local_size: int
+    rank: int
+    node_rank: int
+    local_rank: int
+    device: torch.device
+
+
+def _worker_parallel_launch(
+    local_rank: int,
+    world_size: int,
+    world_local_size: int,
+    node_rank: int,
+    init_method: str,
+    worker: Callable[Concatenate[ProcessGroupInfo, P], None],
+    *args: P.args,
+    **kwargs: P.kwargs,
+) -> None:
+    rank = node_rank * world_local_size + local_rank
+    torch.cuda.set_device(local_rank)
+    device = torch.device("cuda", local_rank)
+    torch.distributed.init_process_group(
+        backend="cpu:gloo,cuda:nccl",
+        init_method=init_method,
+        rank=rank,
+        world_size=world_size,
+        device_id=device,
+    )
+    barrier = torch.tensor([rank], device=device)
+    torch.distributed.all_reduce(barrier)
+
+    try:
+        worker(
+            ProcessGroupInfo(
+                world_size=world_size,
+                world_local_size=world_local_size,
+                rank=rank,
+                node_rank=node_rank,
+                local_rank=local_rank,
+                device=device,
+            ),
+            *args,
+            **kwargs,
+        )
+    except Exception as ex:
+        print(ex)
+        traceback.print_exc()
+        raise
+    finally:
+        torch.distributed.destroy_process_group()
+
+
+def parallel_launch(
+    world_size: int,
+    worker: Callable[Concatenate[ProcessGroupInfo, P], None],
+    *args: P.args,
+    **kwargs: P.kwargs,
+) -> None:
+    assert not kwargs
+    spawn(
+        _worker_parallel_launch,
+        args=(
+            world_size,
+            world_size,
+            0,
+            "tcp://localhost:29500",
+            worker,
+        ) + args,
+        nprocs=world_size,
+        join=True,
+    )
+
+
+def parallel_launch_from_env(
+    worker: Callable[Concatenate[ProcessGroupInfo, P], None],
+    *args: P.args,
+    **kwargs: P.kwargs,
+) -> None:
+    """
+    Launches a worker function in parallel across all processes in the current
+    environment. The environment must have the following variables set:
+    - WORLD_SIZE: The total number of processes.
+    - WORLD_LOCAL_SIZE: The number of processes on the current node.
+    - NODE_RANK: The rank of the current
+    - MASTER_ADDR: The address of the master process.
+    - MASTER_PORT: The port of the master process.
+    """
+    assert not kwargs
+    world_size = int(os.environ["WORLD_SIZE"])
+    world_local_size = int(os.environ["WORLD_LOCAL_SIZE"])
+    node_rank = int(os.environ["NODE_RANK"])
+    assert "MASTER_ADDR" in os.environ
+    assert "MASTER_PORT" in os.environ
+    spawn(
+        _worker_parallel_launch,
+        args=(
+            world_size,
+            world_local_size,
+            node_rank,
+            "env://",
+            worker,
+        ) + args,
+        nprocs=world_local_size,
+        join=True,
+    )

tests/prefix_caching/test_disable_sliding_window.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Compare the with and without prefix caching.
 
 Run `pytest tests/prefix_caching/test_prefix_caching.py`.

tests/prefix_caching/test_prefix_caching.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Compare the with and without prefix caching.
 
 Run `pytest tests/prefix_caching/test_prefix_caching.py`.

tests/prompt_adapter/test_bloom.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import pytest
 

tests/prompt_adapter/test_multi_adapter_inference.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from vllm import EngineArgs, LLMEngine, SamplingParams
 from vllm.prompt_adapter.request import PromptAdapterRequest

tests/prompt_adapter/test_pa_lora.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from huggingface_hub import snapshot_download
 

tests/quantization/test_auto_round.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Test model set-up and inference for quantized HF models supported
  on the AutoRound.
 

tests/quantization/test_bitsandbytes.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 '''Tests whether bitsandbytes computation is enabled correctly.
 
 Run `pytest tests/quantization/test_bitsandbytes.py`.

tests/quantization/test_compressed_tensors.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Test model set-up and weight loading for llmcompressor-quantized models.
 
 Run `pytest tests/quantization/test_compressed_tensors.py`.
@@ -13,9 +14,10 @@ from compressed_tensors.quantization import QuantizationType
 from tests.models.utils import check_logprobs_close
 from vllm.model_executor.layers.quantization.compressed_tensors.compressed_tensors import (  # noqa: E501
     CompressedTensors24, CompressedTensorsLinearMethod,
-    CompressedTensorsW4A16Fp4, CompressedTensorsW4A16Sparse24,
-    CompressedTensorsW8A8Fp8, CompressedTensorsW8A8Int8,
-    CompressedTensorsW8A16Fp8, CompressedTensorsWNA16)
+    CompressedTensorsW4A4Fp4, CompressedTensorsW4A16Fp4,
+    CompressedTensorsW4A16Sparse24, CompressedTensorsW8A8Fp8,
+    CompressedTensorsW8A8Int8, CompressedTensorsW8A16Fp8,
+    CompressedTensorsWNA16)
 from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
     sparse_cutlass_supported)
 from vllm.platforms import current_platform
@@ -650,9 +652,13 @@ def test_compressed_tensors_2of4_sparse_compressed(vllm_runner, args_2of4):
         assert output
 
 
-def test_compressed_tensors_nvfp4a16(vllm_runner):
-    # run weight only example
-    model = "nm-testing/TinyLlama-1.1B-Chat-v1.0-FP4"
+@pytest.mark.parametrize(
+    "args",
+    [("nm-testing/TinyLlama-1.1B-Chat-v1.0-NVFP4A16",
+      CompressedTensorsW4A16Fp4),
+     ("nm-testing/TinyLlama-1.1B-Chat-v1.0-NVFP4", CompressedTensorsW4A4Fp4)])
+def test_compressed_tensors_nvfp4(vllm_runner, args):
+    model, scheme = args
     with vllm_runner(model, enforce_eager=True) as llm:
 
         def check_model(model):
@@ -661,7 +667,7 @@ def test_compressed_tensors_nvfp4a16(vllm_runner):
             qkv_proj = layer.self_attn.qkv_proj
             assert isinstance(qkv_proj.quant_method,
                               CompressedTensorsLinearMethod)
-            assert isinstance(qkv_proj.scheme, CompressedTensorsW4A16Fp4)
+            assert isinstance(qkv_proj.scheme, scheme)
             assert qkv_proj.scheme.group_size == 16
 
         llm.apply_model(check_model)

tests/quantization/test_configs.py

 # SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Tests whether Marlin models can be loaded from the autogptq config.
 
 Run `pytest tests/quantization/test_configs.py --forked`.

tests/quantization/test_cpu_offload.py

-# SPDX-License-Identifier: Apache-2.0
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 # Expanded quantized model tests for CPU offloading
 # Base tests: tests/basic_correctness/test_cpu_offload.py