[misc][plugin] add plugin system implementation (#7426)

.buildkite/test-pipeline.yaml

@@ -77,11 +77,13 @@ steps:
   - pytest -v -s core
 
 - label: Entrypoints Test # 20min
+  working_dir: "/vllm-workspace/tests"
   fast_check: true
   mirror_hardwares: [amd]
   source_file_dependencies:
   - vllm/
   commands:
+  - pip install -e ./plugins/vllm_add_dummy_model
   - pytest -v -s entrypoints/llm
   - pytest -v -s entrypoints/openai
 
@@ -154,6 +156,7 @@ steps:
   - vllm/
   - tests/models
   commands:
+    - pip install -e ./plugins/vllm_add_dummy_model
     - pytest -v -s models -m \"not vlm\"
 
 - label: Vision Language Models Test # 42min
@@ -289,6 +292,7 @@ steps:
   - pytest -v -s distributed/test_chunked_prefill_distributed.py
   - pytest -v -s distributed/test_multimodal_broadcast.py
   - pytest -v -s spec_decode/e2e/test_integration_dist_tp2.py
+  - pytest -v -s distributed/test_distributed_oot.py
   - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s test_sharded_state_loader.py
   - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s distributed/test_utils.py
 

requirements-common.txt

@@ -23,4 +23,5 @@ pyzmq
 librosa # Required for audio processing
 soundfile # Required for audio processing
 gguf == 0.9.1
+importlib_metadata
 compressed-tensors == 0.5.0

tests/conftest.py

@@ -1,7 +1,9 @@
 import contextlib
 import gc
+import json
 import os
 import sys
+import tempfile
 from collections import UserList
 from enum import Enum
 from typing import (Any, Callable, Dict, List, Optional, Tuple, TypedDict,
@@ -11,6 +13,7 @@ import pytest
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
+from huggingface_hub import snapshot_download
 from PIL import Image
 from transformers import (AutoModelForCausalLM, AutoModelForSeq2SeqLM,
                           AutoModelForVision2Seq, AutoTokenizer, BatchEncoding,
@@ -757,3 +760,26 @@ def num_gpus_available():
     in current process."""
 
     return cuda_device_count_stateless()
+
+
+temp_dir = tempfile.gettempdir()
+_dummy_path = os.path.join(temp_dir, "dummy_opt")
+
+
+@pytest.fixture
+def dummy_opt_path():
+    json_path = os.path.join(_dummy_path, "config.json")
+    if not os.path.exists(_dummy_path):
+        snapshot_download(repo_id="facebook/opt-125m",
+                          local_dir=_dummy_path,
+                          ignore_patterns=[
+                              "*.bin", "*.bin.index.json", "*.pt", "*.h5",
+                              "*.msgpack"
+                          ])
+        assert os.path.exists(json_path)
+        with open(json_path, "r") as f:
+            config = json.load(f)
+        config["architectures"] = ["MyOPTForCausalLM"]
+        with open(json_path, "w") as f:
+            json.dump(config, f)
+    return _dummy_path

tests/distributed/test_distributed_oot.py

@@ -0,0 +1,6 @@
+from ..entrypoints.openai.test_oot_registration import (
+    run_and_test_dummy_opt_api_server)
+
+
+def test_distributed_oot(dummy_opt_path: str):
+    run_and_test_dummy_opt_api_server(dummy_opt_path, tp=2)

tests/entrypoints/openai/test_oot_registration.py

@@ -1,70 +1,27 @@
-import sys
-import time
-
-import torch
-from openai import OpenAI, OpenAIError
-
-from vllm import ModelRegistry
-from vllm.model_executor.models.opt import OPTForCausalLM
-from vllm.model_executor.sampling_metadata import SamplingMetadata
-from vllm.utils import get_open_port
-
 from ...utils import VLLM_PATH, RemoteOpenAIServer
 
 chatml_jinja_path = VLLM_PATH / "examples/template_chatml.jinja"
 assert chatml_jinja_path.exists()
 
 
-class MyOPTForCausalLM(OPTForCausalLM):
-
-    def compute_logits(self, hidden_states: torch.Tensor,
-                       sampling_metadata: SamplingMetadata) -> torch.Tensor:
-        # this dummy model always predicts the first token
-        logits = super().compute_logits(hidden_states, sampling_metadata)
-        logits.zero_()
-        logits[:, 0] += 1.0
-        return logits
-
-
-def server_function(port: int):
-    # register our dummy model
-    ModelRegistry.register_model("OPTForCausalLM", MyOPTForCausalLM)
-
-    sys.argv = ["placeholder.py"] + [
-        "--model",
-        "facebook/opt-125m",
+def run_and_test_dummy_opt_api_server(model, tp=1):
+    # the model is registered through the plugin
+    server_args = [
         "--gpu-memory-utilization",
         "0.10",
         "--dtype",
         "float32",
-        "--api-key",
-        "token-abc123",
-        "--port",
-        str(port),
         "--chat-template",
         str(chatml_jinja_path),
+        "--load-format",
+        "dummy",
+        "-tp",
+        f"{tp}",
     ]
-
-    import runpy
-    runpy.run_module('vllm.entrypoints.openai.api_server', run_name='__main__')
-
-
-def test_oot_registration_for_api_server():
-    port = get_open_port()
-    ctx = torch.multiprocessing.get_context()
-    server = ctx.Process(target=server_function, args=(port, ))
-    server.start()
-
-    try:
-        client = OpenAI(
-            base_url=f"http://localhost:{port}/v1",
-            api_key="token-abc123",
-        )
-        now = time.time()
-        while True:
-            try:
+    with RemoteOpenAIServer(model, server_args) as server:
+        client = server.get_client()
         completion = client.chat.completions.create(
-                    model="facebook/opt-125m",
+            model=model,
             messages=[{
                 "role": "system",
                 "content": "You are a helpful assistant."
@@ -74,21 +31,12 @@ def test_oot_registration_for_api_server():
             }],
             temperature=0,
         )
-                break
-            except OpenAIError as e:
-                if "Connection error" in str(e):
-                    time.sleep(3)
-                    if time.time() - now > RemoteOpenAIServer.MAX_START_WAIT_S:
-                        msg = "Server did not start in time"
-                        raise RuntimeError(msg) from e
-                else:
-                    raise e
-    finally:
-        server.terminate()
-
         generated_text = completion.choices[0].message.content
         assert generated_text is not None
         # make sure only the first token is generated
-    # TODO(youkaichao): Fix the test with plugin
-    rest = generated_text.replace("<s>", "")  # noqa
-    # assert rest == ""
+        rest = generated_text.replace("<s>", "")
+        assert rest == ""
+
+
+def test_oot_registration_for_api_server(dummy_opt_path: str):
+    run_and_test_dummy_opt_api_server(dummy_opt_path)

tests/models/test_oot_registration.py

@@ -1,32 +1,27 @@
-from typing import Optional
+import os
 
-import torch
+import pytest
 
-from vllm import LLM, ModelRegistry, SamplingParams
-from vllm.model_executor.models.opt import OPTForCausalLM
-from vllm.model_executor.sampling_metadata import SamplingMetadata
+from vllm import LLM, SamplingParams
+
+# NOTE: the order of the tests is important
+# the first test does not load any plugins
+# the second test loads the plugin
+# they share the same process, so the plugin is loaded for the second test
 
 
-class MyOPTForCausalLM(OPTForCausalLM):
-
-    def compute_logits(
-        self,
-        hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
-    ) -> Optional[torch.Tensor]:
-        # this dummy model always predicts the first token
-        logits = super().compute_logits(hidden_states, sampling_metadata)
-        logits.zero_()
-        logits[:, 0] += 1.0
-        return logits
+def test_plugin(dummy_opt_path):
+    os.environ["VLLM_PLUGINS"] = ""
+    with pytest.raises(Exception) as excinfo:
+        LLM(model=dummy_opt_path, load_format="dummy")
+    assert "are not supported for now" in str(excinfo.value)
 
 
-def test_oot_registration():
-    # register our dummy model
-    ModelRegistry.register_model("OPTForCausalLM", MyOPTForCausalLM)
+def test_oot_registration(dummy_opt_path):
+    os.environ["VLLM_PLUGINS"] = "register_dummy_model"
     prompts = ["Hello, my name is", "The text does not matter"]
     sampling_params = SamplingParams(temperature=0)
-    llm = LLM(model="facebook/opt-125m")
+    llm = LLM(model=dummy_opt_path, load_format="dummy")
     first_token = llm.get_tokenizer().decode(0)
     outputs = llm.generate(prompts, sampling_params)
 

tests/plugins/vllm_add_dummy_model/setup.py

@@ -0,0 +1,9 @@
+from setuptools import setup
+
+setup(name='vllm_add_dummy_model',
+      version='0.1',
+      packages=['vllm_add_dummy_model'],
+      entry_points={
+          'vllm.general_plugins':
+          ["register_dummy_model = vllm_add_dummy_model:register"]
+      })

tests/plugins/vllm_add_dummy_model/vllm_add_dummy_model/__init__.py

@@ -0,0 +1,26 @@
+from typing import Optional
+
+import torch
+
+from vllm import ModelRegistry
+from vllm.model_executor.models.opt import OPTForCausalLM
+from vllm.model_executor.sampling_metadata import SamplingMetadata
+
+
+class MyOPTForCausalLM(OPTForCausalLM):
+
+    def compute_logits(
+            self, hidden_states: torch.Tensor,
+            sampling_metadata: SamplingMetadata) -> Optional[torch.Tensor]:
+        # this dummy model always predicts the first token
+        logits = super().compute_logits(hidden_states, sampling_metadata)
+        if logits is not None:
+            logits.zero_()
+            logits[:, 0] += 1.0
+        return logits
+
+
+def register():
+    # register our dummy model
+    if "MyOPTForCausalLM" not in ModelRegistry.get_supported_archs():
+        ModelRegistry.register_model("MyOPTForCausalLM", MyOPTForCausalLM)

vllm/engine/llm_engine.py

@@ -227,6 +227,9 @@ class LLMEngine:
         )
         # TODO(woosuk): Print more configs in debug mode.
 
+        from vllm.plugins import load_general_plugins
+        load_general_plugins()
+
         self.model_config = model_config
         self.cache_config = cache_config
         self.lora_config = lora_config

vllm/envs.py

@@ -1,6 +1,6 @@
 import os
 import tempfile
-from typing import TYPE_CHECKING, Any, Callable, Dict, Optional
+from typing import TYPE_CHECKING, Any, Callable, Dict, List, Optional
 
 if TYPE_CHECKING:
     VLLM_HOST_IP: str = ""
@@ -55,6 +55,7 @@ if TYPE_CHECKING:
     VERBOSE: bool = False
     VLLM_ALLOW_LONG_MAX_MODEL_LEN: bool = False
     VLLM_TEST_FORCE_FP8_MARLIN: bool = False
+    VLLM_PLUGINS: Optional[List[str]] = None
 
 
 def get_default_cache_root():
@@ -362,6 +363,13 @@ environment_variables: Dict[str, Callable[[], Any]] = {
     lambda:
     (os.environ.get("VLLM_TEST_FORCE_FP8_MARLIN", "0").strip().lower() in
      ("1", "true")),
+
+    # a list of plugin names to load, separated by commas.
+    # if this is not set, it means all plugins will be loaded
+    # if this is set to an empty string, no plugins will be loaded
+    "VLLM_PLUGINS":
+    lambda: None if "VLLM_PLUGINS" not in os.environ else os.environ[
+        "VLLM_PLUGINS"].split(","),
 }
 
 # end-env-vars-definition

vllm/model_executor/models/__init__.py

@@ -166,7 +166,7 @@ class ModelRegistry:
 
     @staticmethod
     def get_supported_archs() -> List[str]:
-        return list(_MODELS.keys())
+        return list(_MODELS.keys()) + list(_OOT_MODELS.keys())
 
     @staticmethod
     def register_model(model_arch: str, model_cls: Type[nn.Module]):

vllm/plugins/__init__.py

@@ -0,0 +1,31 @@
+import logging
+
+import vllm.envs as envs
+
+logger = logging.getLogger(__name__)
+
+
+def load_general_plugins():
+    """WARNING: plugins can be loaded for multiple times in different
+    processes. They should be designed in a way that they can be loaded
+    multiple times without causing issues.
+    """
+    import sys
+    if sys.version_info < (3, 10):
+        from importlib_metadata import entry_points
+    else:
+        from importlib.metadata import entry_points
+
+    allowed_plugins = envs.VLLM_PLUGINS
+
+    discovered_plugins = entry_points(group='vllm.general_plugins')
+    for plugin in discovered_plugins:
+        logger.info("Found general plugin: %s", plugin.name)
+        if allowed_plugins is None or plugin.name in allowed_plugins:
+            try:
+                func = plugin.load()
+                func()
+                logger.info("Loaded general plugin: %s", plugin.name)
+            except Exception:
+                logger.exception("Failed to load general plugin: %s",
+                                 plugin.name)

vllm/worker/worker_base.py

@@ -411,6 +411,9 @@ class WorkerWrapperBase:
         # see https://github.com/NVIDIA/nccl/issues/1234
         os.environ['NCCL_CUMEM_ENABLE'] = '0'
 
+        from vllm.plugins import load_general_plugins
+        load_general_plugins()
+
         if self.worker_class_fn:
             worker_class = self.worker_class_fn()
         else: