Add vLLM v0.18.1 source tree with KV transfer abort fix

third_party/vllm/ now tracked in git for direct patch management.
Based on vLLM v0.18.1 release with one patch applied:

  vllm/v1/core/sched/scheduler.py:
    Replace fatal assert with graceful skip when KV transfer callback
    arrives for an already-aborted request during PD disaggregated serving.

Future vLLM modifications should be made directly in third_party/vllm/
and committed normally. The patches/ directory is kept as documentation
of what changed from upstream.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

third_party/vllm/tests/entrypoints/pooling/classify/test_offline.py

@@ -0,0 +1,114 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import weakref
+
+import pytest
+import torch
+
+from tests.models.utils import softmax
+from vllm import LLM, ClassificationRequestOutput, PoolingParams, PoolingRequestOutput
+from vllm.distributed import cleanup_dist_env_and_memory
+from vllm.tasks import PoolingTask
+
+MODEL_NAME = "jason9693/Qwen2.5-1.5B-apeach"
+
+prompt = "The chef prepared a delicious meal."
+prompt_token_ids = [785, 29706, 10030, 264, 17923, 15145, 13]
+num_labels = 2
+
+
+@pytest.fixture(scope="module")
+def llm():
+    # pytest caches the fixture so we use weakref.proxy to
+    # enable garbage collection
+    llm = LLM(
+        model=MODEL_NAME,
+        max_num_batched_tokens=32768,
+        tensor_parallel_size=1,
+        gpu_memory_utilization=0.75,
+        enforce_eager=True,
+        seed=0,
+    )
+
+    yield weakref.proxy(llm)
+
+    del llm
+
+    cleanup_dist_env_and_memory()
+
+
+@pytest.mark.skip_global_cleanup
+def test_str_prompts(llm: LLM):
+    outputs = llm.classify(prompt, use_tqdm=False)
+    assert len(outputs) == 1
+    assert isinstance(outputs[0], ClassificationRequestOutput)
+    assert outputs[0].prompt_token_ids == prompt_token_ids
+    assert len(outputs[0].outputs.probs) == num_labels
+
+
+@pytest.mark.skip_global_cleanup
+def test_token_ids_prompts(llm: LLM):
+    outputs = llm.classify([prompt_token_ids], use_tqdm=False)
+    assert len(outputs) == 1
+    assert isinstance(outputs[0], ClassificationRequestOutput)
+    assert outputs[0].prompt_token_ids == prompt_token_ids
+    assert len(outputs[0].outputs.probs) == num_labels
+
+
+@pytest.mark.skip_global_cleanup
+def test_list_prompts(llm: LLM):
+    outputs = llm.classify([prompt, prompt_token_ids], use_tqdm=False)
+    assert len(outputs) == 2
+    for i in range(len(outputs)):
+        assert isinstance(outputs[i], ClassificationRequestOutput)
+        assert outputs[i].prompt_token_ids == prompt_token_ids
+        assert len(outputs[i].outputs.probs) == num_labels
+
+
+@pytest.mark.skip_global_cleanup
+def test_token_classify(llm: LLM):
+    outputs = llm.encode(prompt, pooling_task="token_classify", use_tqdm=False)
+    assert len(outputs) == 1
+    assert isinstance(outputs[0], PoolingRequestOutput)
+    assert outputs[0].prompt_token_ids == prompt_token_ids
+    assert outputs[0].outputs.data.shape == (len(prompt_token_ids), num_labels)
+
+
+@pytest.mark.skip_global_cleanup
+def test_pooling_params(llm: LLM):
+    def get_outputs(use_activation):
+        outputs = llm.classify(
+            prompt,
+            pooling_params=PoolingParams(use_activation=use_activation),
+            use_tqdm=False,
+        )
+        return torch.tensor([x.outputs.probs for x in outputs])
+
+    default = get_outputs(use_activation=None)
+    w_activation = get_outputs(use_activation=True)
+    wo_activation = get_outputs(use_activation=False)
+
+    assert torch.allclose(default, w_activation, atol=1e-2), (
+        "Default should use activation."
+    )
+    assert not torch.allclose(w_activation, wo_activation, atol=1e-2), (
+        "wo_activation should not use activation."
+    )
+    assert torch.allclose(softmax(wo_activation), w_activation, atol=1e-2), (
+        "w_activation should be close to activation(wo_activation)."
+    )
+
+
+@pytest.mark.skip_global_cleanup
+def test_score_api(llm: LLM):
+    err_msg = "Score API is only enabled for num_labels == 1."
+    with pytest.raises(ValueError, match=err_msg):
+        llm.score("ping", "pong", use_tqdm=False)
+
+
+@pytest.mark.parametrize("task", ["embed", "token_embed", "plugin"])
+def test_unsupported_tasks(llm: LLM, task: PoolingTask):
+    err_msg = f"Unsupported task: '{task}' Supported tasks.+"
+    with pytest.raises(ValueError, match=err_msg):
+        llm.encode(prompt, pooling_task=task, use_tqdm=False)

third_party/vllm/tests/entrypoints/pooling/classify/test_online.py

@@ -0,0 +1,472 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import pytest
+import requests
+import torch
+import torch.nn.functional as F
+
+from tests.utils import RemoteOpenAIServer
+from vllm.entrypoints.pooling.classify.protocol import ClassificationResponse
+from vllm.entrypoints.pooling.pooling.protocol import PoolingResponse
+
+MODEL_NAME = "jason9693/Qwen2.5-1.5B-apeach"
+DTYPE = "float32"  # Use float32 to avoid NaN issue
+input_text = "This product was excellent and exceeded my expectations"
+input_tokens = [1986, 1985, 572, 9073, 323, 33808, 847, 16665]
+
+
+@pytest.fixture(scope="module")
+def server():
+    args = [
+        "--enforce-eager",
+        "--max-model-len",
+        "512",
+        "--dtype",
+        DTYPE,
+    ]
+
+    with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
+        yield remote_server
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_basic(server: RemoteOpenAIServer, model_name: str):
+    # test /v1/models
+    response = requests.get(server.url_for("/v1/models"))
+    served_model = response.json()["data"][0]["id"]
+    assert served_model == MODEL_NAME
+
+    # test /tokenize
+    response = requests.post(
+        server.url_for("/tokenize"),
+        json={"model": model_name, "prompt": input_text},
+    )
+    assert response.json()["tokens"] == input_tokens
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_completion_request(server: RemoteOpenAIServer, model_name: str):
+    # test input: str
+    classification_response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": input_text},
+    )
+
+    classification_response.raise_for_status()
+    output = ClassificationResponse.model_validate(classification_response.json())
+
+    assert output.object == "list"
+    assert output.model == MODEL_NAME
+    assert len(output.data) == 1
+    assert hasattr(output.data[0], "label")
+    assert hasattr(output.data[0], "probs")
+
+    # test input: list[int]
+    classification_response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": input_tokens},
+    )
+
+    classification_response.raise_for_status()
+    output = ClassificationResponse.model_validate(classification_response.json())
+
+    assert output.object == "list"
+    assert output.model == MODEL_NAME
+    assert len(output.data) == 1
+    assert hasattr(output.data[0], "label")
+    assert hasattr(output.data[0], "probs")
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_completion_request_batched(server: RemoteOpenAIServer, model_name: str):
+    N = 10
+
+    # test input: list[str]
+    classification_response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": [input_text] * N},
+    )
+    output = ClassificationResponse.model_validate(classification_response.json())
+
+    assert len(output.data) == N
+    for i, item in enumerate(output.data):
+        assert item.index == i
+        assert hasattr(item, "label")
+        assert hasattr(item, "probs")
+        assert len(item.probs) == item.num_classes
+        assert item.label in ["Default", "Spoiled"]
+
+    # test input: list[list[int]]
+    classification_response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": [input_tokens] * N},
+    )
+    output = ClassificationResponse.model_validate(classification_response.json())
+
+    assert len(output.data) == N
+    for i, item in enumerate(output.data):
+        assert item.index == i
+        assert hasattr(item, "label")
+        assert hasattr(item, "probs")
+        assert len(item.probs) == item.num_classes
+        assert item.label in ["Default", "Spoiled"]
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_empty_input_error(server: RemoteOpenAIServer, model_name: str):
+    classification_response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": ""},
+    )
+
+    error = classification_response.json()
+    assert classification_response.status_code == 400
+    assert "error" in error
+
+    classification_response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": []},
+    )
+
+    error = classification_response.json()
+    assert classification_response.status_code == 400
+    assert "error" in error
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_truncate_prompt_tokens(server: RemoteOpenAIServer, model_name: str):
+    long_text = "hello " * 600
+
+    classification_response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": long_text, "truncate_prompt_tokens": 5},
+    )
+
+    classification_response.raise_for_status()
+    output = ClassificationResponse.model_validate(classification_response.json())
+
+    assert len(output.data) == 1
+    assert output.data[0].index == 0
+    assert hasattr(output.data[0], "probs")
+    assert output.usage.prompt_tokens == 5
+    assert output.usage.total_tokens == 5
+
+    # invalid_truncate_prompt_tokens
+    classification_response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": "test", "truncate_prompt_tokens": 513},
+    )
+
+    error = classification_response.json()
+    assert classification_response.status_code == 400
+    assert "truncate_prompt_tokens" in error["error"]["message"]
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_add_special_tokens(server: RemoteOpenAIServer, model_name: str):
+    # The add_special_tokens parameter doesn't seem to be working with this model.
+    # working with papluca/xlm-roberta-base-language-detection
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": input_text, "add_special_tokens": False},
+    )
+    response.raise_for_status()
+    ClassificationResponse.model_validate(response.json())
+
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "input": input_text, "add_special_tokens": True},
+    )
+    response.raise_for_status()
+    ClassificationResponse.model_validate(response.json())
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+async def test_chat_request(server: RemoteOpenAIServer, model_name: str):
+    messages = [
+        {
+            "role": "user",
+            "content": "The cat sat on the mat.",
+        },
+        {
+            "role": "assistant",
+            "content": "A feline was resting on a rug.",
+        },
+        {
+            "role": "user",
+            "content": "Stars twinkle brightly in the night sky.",
+        },
+    ]
+
+    # test chat request basic usage
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "messages": messages},
+    )
+
+    response.raise_for_status()
+    output = ClassificationResponse.model_validate(response.json())
+
+    assert output.object == "list"
+    assert output.model == MODEL_NAME
+    assert len(output.data) == 1
+    assert hasattr(output.data[0], "label")
+    assert hasattr(output.data[0], "probs")
+    assert output.usage.prompt_tokens == 51
+
+    # test add_generation_prompt
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "messages": messages, "add_generation_prompt": True},
+    )
+
+    response.raise_for_status()
+    output = ClassificationResponse.model_validate(response.json())
+
+    assert output.object == "list"
+    assert output.model == MODEL_NAME
+    assert len(output.data) == 1
+    assert hasattr(output.data[0], "label")
+    assert hasattr(output.data[0], "probs")
+    assert output.usage.prompt_tokens == 54
+
+    # test continue_final_message
+    response = requests.post(
+        server.url_for("classify"),
+        json={
+            "model": model_name,
+            "messages": messages,
+            "continue_final_message": True,
+        },
+    )
+
+    response.raise_for_status()
+    output = ClassificationResponse.model_validate(response.json())
+
+    assert output.object == "list"
+    assert output.model == MODEL_NAME
+    assert len(output.data) == 1
+    assert hasattr(output.data[0], "label")
+    assert hasattr(output.data[0], "probs")
+    assert output.usage.prompt_tokens == 49
+
+    # test add_special_tokens
+    # The add_special_tokens parameter doesn't seem to be working with this model.
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "messages": messages, "add_special_tokens": True},
+    )
+
+    response.raise_for_status()
+    output = ClassificationResponse.model_validate(response.json())
+
+    assert output.object == "list"
+    assert output.model == MODEL_NAME
+    assert len(output.data) == 1
+    assert hasattr(output.data[0], "label")
+    assert hasattr(output.data[0], "probs")
+    assert output.usage.prompt_tokens == 51
+
+    # test continue_final_message with add_generation_prompt
+    response = requests.post(
+        server.url_for("classify"),
+        json={
+            "model": model_name,
+            "messages": messages,
+            "continue_final_message": True,
+            "add_generation_prompt": True,
+        },
+    )
+    assert (
+        "Cannot set both `continue_final_message` and `add_generation_prompt` to True."
+        in response.json()["error"]["message"]
+    )
+
+
+@pytest.mark.asyncio
+async def test_invocations_completion_request(server: RemoteOpenAIServer):
+    request_args = {
+        "model": MODEL_NAME,
+        "input": input_text,
+    }
+
+    classification_response = requests.post(
+        server.url_for("classify"), json=request_args
+    )
+    classification_response.raise_for_status()
+
+    invocation_response = requests.post(
+        server.url_for("invocations"), json=request_args
+    )
+    invocation_response.raise_for_status()
+
+    classification_output = classification_response.json()
+    invocation_output = invocation_response.json()
+
+    assert classification_output.keys() == invocation_output.keys()
+    for classification_data, invocation_data in zip(
+        classification_output["data"], invocation_output["data"]
+    ):
+        assert classification_data.keys() == invocation_data.keys()
+        assert classification_data["probs"] == pytest.approx(
+            invocation_data["probs"], rel=0.01
+        )
+
+
+@pytest.mark.asyncio
+async def test_invocations_chat_request(server: RemoteOpenAIServer):
+    messages = [
+        {
+            "role": "user",
+            "content": "The cat sat on the mat.",
+        },
+        {
+            "role": "assistant",
+            "content": "A feline was resting on a rug.",
+        },
+        {
+            "role": "user",
+            "content": "Stars twinkle brightly in the night sky.",
+        },
+    ]
+
+    request_args = {"model": MODEL_NAME, "messages": messages}
+
+    classification_response = requests.post(
+        server.url_for("classify"), json=request_args
+    )
+    classification_response.raise_for_status()
+
+    invocation_response = requests.post(
+        server.url_for("invocations"), json=request_args
+    )
+    invocation_response.raise_for_status()
+
+    classification_output = classification_response.json()
+    invocation_output = invocation_response.json()
+
+    assert classification_output.keys() == invocation_output.keys()
+    for classification_data, invocation_data in zip(
+        classification_output["data"], invocation_output["data"]
+    ):
+        assert classification_data.keys() == invocation_data.keys()
+        assert classification_data["probs"] == pytest.approx(
+            invocation_data["probs"], rel=0.01
+        )
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+async def test_use_activation(server: RemoteOpenAIServer, model_name: str):
+    async def get_outputs(use_activation):
+        response = requests.post(
+            server.url_for("classify"),
+            json={
+                "model": model_name,
+                "input": input_text,
+                "use_activation": use_activation,
+            },
+        )
+        outputs = response.json()
+        return torch.tensor([x["probs"] for x in outputs["data"]])
+
+    default = await get_outputs(use_activation=None)
+    w_activation = await get_outputs(use_activation=True)
+    wo_activation = await get_outputs(use_activation=False)
+
+    assert torch.allclose(default, w_activation, atol=1e-2), (
+        "Default should use activation."
+    )
+    assert not torch.allclose(w_activation, wo_activation, atol=1e-2), (
+        "wo_activation should not use activation."
+    )
+    assert torch.allclose(F.softmax(wo_activation, dim=-1), w_activation, atol=1e-2), (
+        "w_activation should be close to activation(wo_activation)."
+    )
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+async def test_score(server: RemoteOpenAIServer, model_name: str):
+    # score api is only enabled for num_labels == 1.
+    response = requests.post(
+        server.url_for("score"),
+        json={
+            "model": model_name,
+            "queries": "ping",
+            "documents": "pong",
+        },
+    )
+    assert response.json()["detail"] == "Not Found"
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+async def test_rerank(server: RemoteOpenAIServer, model_name: str):
+    # rerank api is only enabled for num_labels == 1.
+    response = requests.post(
+        server.url_for("rerank"),
+        json={
+            "model": model_name,
+            "query": "ping",
+            "documents": ["pong"],
+        },
+    )
+    assert response.json()["detail"] == "Not Found"
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+async def test_pooling_classify(server: RemoteOpenAIServer, model_name: str):
+    response = requests.post(
+        server.url_for("pooling"),
+        json={
+            "model": model_name,
+            "input": input_text,
+            "encoding_format": "float",
+            "task": "classify",
+        },
+    )
+    poolings = PoolingResponse.model_validate(response.json())
+    assert len(poolings.data) == 1
+    assert len(poolings.data[0].data) == 2
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+async def test_pooling_token_classify(server: RemoteOpenAIServer, model_name: str):
+    task = "token_classify"
+    response = requests.post(
+        server.url_for("pooling"),
+        json={
+            "model": model_name,
+            "input": input_text,
+            "encoding_format": "float",
+            "task": task,
+        },
+    )
+    poolings = PoolingResponse.model_validate(response.json())
+    assert len(poolings.data) == 1
+    assert len(poolings.data[0].data) == 8
+    assert len(poolings.data[0].data[0]) == 2
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+@pytest.mark.parametrize("task", ["embed", "token_embed", "plugin"])
+async def test_pooling_not_supported(
+    server: RemoteOpenAIServer, model_name: str, task: str
+):
+    response = requests.post(
+        server.url_for("pooling"),
+        json={
+            "model": model_name,
+            "input": input_text,
+            "encoding_format": "float",
+            "task": task,
+        },
+    )
+    assert response.json()["error"]["type"] == "BadRequestError"
+    assert response.json()["error"]["message"].startswith(f"Unsupported task: {task!r}")

third_party/vllm/tests/entrypoints/pooling/classify/test_online_vision.py

@@ -0,0 +1,146 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+import json
+
+import pytest
+import requests
+
+from tests.utils import RemoteOpenAIServer
+from vllm.entrypoints.pooling.classify.protocol import ClassificationResponse
+from vllm.multimodal.utils import encode_image_url, fetch_image
+
+MODEL_NAME = "muziyongshixin/Qwen2.5-VL-7B-for-VideoCls"
+MAXIMUM_VIDEOS = 1
+
+HF_OVERRIDES = {"architectures": ["Qwen2_5_VLForSequenceClassification"]}
+input_text = "This product was excellent and exceeded my expectations"
+image_url = "https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/cat_snow.jpg"
+image_base64 = {"url": encode_image_url(fetch_image(image_url))}
+video_url = "https://www.bogotobogo.com/python/OpenCV_Python/images/mean_shift_tracking/slow_traffic_small.mp4"
+
+
+@pytest.fixture(scope="module")
+def server():
+    args = [
+        "--runner",
+        "pooling",
+        "--max-model-len",
+        "5000",
+        "--enforce-eager",
+        "--limit-mm-per-prompt",
+        json.dumps({"video": MAXIMUM_VIDEOS}),
+    ]
+
+    with RemoteOpenAIServer(
+        MODEL_NAME, args, override_hf_configs=HF_OVERRIDES
+    ) as remote_server:
+        yield remote_server
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_chat_text_request(server: RemoteOpenAIServer, model_name: str):
+    messages = [
+        {
+            "role": "assistant",
+            "content": "Please classify this text request.",
+        },
+        {
+            "role": "user",
+            "content": input_text,
+        },
+    ]
+
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "messages": messages},
+    )
+    response.raise_for_status()
+
+    output = ClassificationResponse.model_validate(response.json())
+
+    assert output.object == "list"
+    assert output.model == model_name
+    assert len(output.data) == 1
+    assert len(output.data[0].probs) == 2
+    assert output.usage.prompt_tokens == 35
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_chat_image_url_request(server: RemoteOpenAIServer, model_name: str):
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "Please classify this image."},
+                {"type": "image_url", "image_url": {"url": image_url}},
+            ],
+        }
+    ]
+
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "messages": messages},
+    )
+    response.raise_for_status()
+
+    output = ClassificationResponse.model_validate(response.json())
+
+    assert output.object == "list"
+    assert output.model == model_name
+    assert len(output.data) == 1
+    assert len(output.data[0].probs) == 2
+    assert output.usage.prompt_tokens == 47
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_chat_image_base64_request(server: RemoteOpenAIServer, model_name: str):
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "Please classify this image."},
+                {"type": "image_url", "image_url": image_base64},
+            ],
+        }
+    ]
+
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "messages": messages},
+    )
+    response.raise_for_status()
+
+    output = ClassificationResponse.model_validate(response.json())
+
+    assert output.object == "list"
+    assert output.model == model_name
+    assert len(output.data) == 1
+    assert len(output.data[0].probs) == 2
+    assert output.usage.prompt_tokens == 47
+
+
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+def test_chat_video_url_request(server: RemoteOpenAIServer, model_name: str):
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "Please classify this video."},
+                {"type": "video_url", "video_url": {"url": video_url}},
+            ],
+        }
+    ]
+
+    response = requests.post(
+        server.url_for("classify"),
+        json={"model": model_name, "messages": messages},
+    )
+    response.raise_for_status()
+
+    output = ClassificationResponse.model_validate(response.json())
+
+    assert output.object == "list"
+    assert output.model == model_name
+    assert len(output.data) == 1
+    assert len(output.data[0].probs) == 2
+    assert output.usage.prompt_tokens == 4807