[Misc] Clean up more utils

requirements/docs.txt

@@ -13,6 +13,8 @@ ruff
 # Required for argparse hook only
 -f https://download.pytorch.org/whl/cpu
 cachetools
+cloudpickle
+py-cpuinfo
 msgspec
 pydantic
 torch

tools/pre_commit/check_pickle_imports.py

@@ -39,7 +39,6 @@
     "vllm/v1/executor/multiproc_executor.py",
     "vllm/v1/executor/ray_executor.py",
     "vllm/entrypoints/llm.py",
-    "vllm/utils/__init__.py",
     "tests/utils.py",
     # pickle and cloudpickle
     "vllm/v1/serial_utils.py",

vllm/config/model.py

@@ -1618,6 +1618,29 @@ def is_encoder_decoder(self) -> bool:
         """Extract the HF encoder/decoder model flag."""
         return is_encoder_decoder(self.hf_config)
 
+    @property
+    def uses_alibi(self) -> bool:
+        cfg = self.hf_text_config
+
+        return (
+            getattr(cfg, "alibi", False)  # Falcon
+            or "BloomForCausalLM" in self.architectures  # Bloom
+            or getattr(cfg, "position_encoding_type", "") == "alibi"  # codellm_1b_alibi
+            or (
+                hasattr(cfg, "attn_config")  # MPT
+                and (
+                    (
+                        isinstance(cfg.attn_config, dict)
+                        and cfg.attn_config.get("alibi", False)
+                    )
+                    or (
+                        not isinstance(cfg.attn_config, dict)
+                        and getattr(cfg.attn_config, "alibi", False)
+                    )
+                )
+            )
+        )
+
     @property
     def uses_mrope(self) -> bool:
         return uses_mrope(self.hf_config)

vllm/config/vllm.py

@@ -2,12 +2,16 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import copy
+import getpass
 import hashlib
 import json
 import os
+import tempfile
+import threading
 import time
 from contextlib import contextmanager
 from dataclasses import replace
+from datetime import datetime
 from functools import lru_cache
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, TypeVar
@@ -17,7 +21,7 @@
 from pydantic.dataclasses import dataclass
 
 import vllm.envs as envs
-from vllm.logger import init_logger
+from vllm.logger import enable_trace_function_call, init_logger
 from vllm.transformers_utils.runai_utils import is_runai_obj_uri
 from vllm.utils import random_uuid
 
@@ -206,6 +210,28 @@ def pad_for_cudagraph(self, batch_size: int) -> int:
         # i.e., batch_size <= self.compilation_config.max_cudagraph_capture_size
         return self.compilation_config.bs_to_padded_graph_size[batch_size]
 
+    def enable_trace_function_call_for_thread(self) -> None:
+        """
+        Set up function tracing for the current thread,
+        if enabled via the `VLLM_TRACE_FUNCTION` environment variable.
+        """
+        if envs.VLLM_TRACE_FUNCTION:
+            tmp_dir = tempfile.gettempdir()
+            # add username to tmp_dir to avoid permission issues
+            tmp_dir = os.path.join(tmp_dir, getpass.getuser())
+            filename = (
+                f"VLLM_TRACE_FUNCTION_for_process_{os.getpid()}"
+                f"_thread_{threading.get_ident()}_at_{datetime.now()}.log"
+            ).replace(" ", "_")
+            log_path = os.path.join(
+                tmp_dir,
+                "vllm",
+                f"vllm-instance-{self.instance_id}",
+                filename,
+            )
+            os.makedirs(os.path.dirname(log_path), exist_ok=True)
+            enable_trace_function_call(log_path)
+
     @staticmethod
     def _get_quantization_config(
         model_config: ModelConfig, load_config: LoadConfig

vllm/engine/arg_utils.py

@@ -73,7 +73,7 @@
 from vllm.logger import init_logger
 from vllm.platforms import CpuArchEnum, current_platform
 from vllm.plugins import load_general_plugins
-from vllm.ray.lazy_utils import is_ray_initialized
+from vllm.ray.lazy_utils import is_in_ray_actor, is_ray_initialized
 from vllm.reasoning import ReasoningParserManager
 from vllm.test_utils import MODEL_WEIGHTS_S3_BUCKET, MODELS_ON_S3
 from vllm.transformers_utils.config import (
@@ -82,7 +82,6 @@
     maybe_override_with_speculators,
 )
 from vllm.transformers_utils.utils import check_gguf_file
-from vllm.utils import is_in_ray_actor
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 from vllm.utils.mem_constants import GiB_bytes
 from vllm.utils.network_utils import get_ip

vllm/entrypoints/anthropic/api_server.py

@@ -51,9 +51,9 @@
     with_cancellation,
 )
 from vllm.logger import init_logger
-from vllm.utils import set_ulimit
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 from vllm.utils.network_utils import is_valid_ipv6_address
+from vllm.utils.system_utils import set_ulimit
 from vllm.version import __version__ as VLLM_VERSION
 
 prometheus_multiproc_dir: tempfile.TemporaryDirectory

vllm/entrypoints/api_server.py

@@ -26,8 +26,9 @@
 from vllm.logger import init_logger
 from vllm.sampling_params import SamplingParams
 from vllm.usage.usage_lib import UsageContext
-from vllm.utils import random_uuid, set_ulimit
+from vllm.utils import random_uuid
 from vllm.utils.argparse_utils import FlexibleArgumentParser
+from vllm.utils.system_utils import set_ulimit
 from vllm.version import __version__ as VLLM_VERSION
 
 logger = init_logger("vllm.entrypoints.api_server")

vllm/entrypoints/openai/api_server.py

@@ -108,10 +108,10 @@
 from vllm.logger import init_logger
 from vllm.reasoning import ReasoningParserManager
 from vllm.usage.usage_lib import UsageContext
-from vllm.utils import Device, set_ulimit
+from vllm.utils import Device
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 from vllm.utils.network_utils import is_valid_ipv6_address
-from vllm.utils.system_utils import decorate_logs
+from vllm.utils.system_utils import decorate_logs, set_ulimit
 from vllm.v1.engine.exceptions import EngineDeadError
 from vllm.v1.metrics.prometheus import get_prometheus_registry
 from vllm.version import __version__ as VLLM_VERSION

vllm/platforms/__init__.py

@@ -60,7 +60,7 @@ def cuda_platform_plugin() -> str | None:
     is_cuda = False
     logger.debug("Checking if CUDA platform is available.")
     try:
-        from vllm.utils import import_pynvml
+        from vllm.utils.import_utils import import_pynvml
 
         pynvml = import_pynvml()
         pynvml.nvmlInit()

vllm/platforms/cuda.py

@@ -16,7 +16,7 @@
 import vllm._C  # noqa
 import vllm.envs as envs
 from vllm.logger import init_logger
-from vllm.utils import import_pynvml
+from vllm.utils.import_utils import import_pynvml
 from vllm.utils.torch_utils import cuda_device_count_stateless
 
 from .interface import DeviceCapability, Platform, PlatformEnum