A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 02:15 UTC
[Deprecation] Remove `TokenizerPoolConfig` (#20968)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Jul 15, 2025 at 15:00 UTC
56fe4bedd6df401037a251b2b986767a290ad2e7
4 files changed
+4
-62
docs/api/README.md
-1
@@ -8,7 +8,6 @@ API documentation for vLLM's configuration classes.
8
9
- [vllm.config.ModelConfig][]
10
- [vllm.config.CacheConfig][]
11
-- [vllm.config.TokenizerPoolConfig][]
11
- [vllm.config.LoadConfig][]
12
- [vllm.config.ParallelConfig][]
13
- [vllm.config.SchedulerConfig][]
tests/async_engine/test_api_server.py
+2
-6
@@ -29,7 +29,7 @@ def _query_server_long(prompt: str) -> dict:
29
30
31
@pytest.fixture
32
-def api_server(tokenizer_pool_size: int, distributed_executor_backend: str):
32
+def api_server(distributed_executor_backend: str):
33
script_path = Path(__file__).parent.joinpath(
34
"api_server_async_engine.py").absolute()
35
commands = [
@@ -40,8 +40,6 @@ def api_server(tokenizer_pool_size: int, distributed_executor_backend: str):
40
"facebook/opt-125m",
41
"--host",
42
"127.0.0.1",
43
- "--tokenizer-pool-size",
44
- str(tokenizer_pool_size),
43
"--distributed-executor-backend",
44
distributed_executor_backend,
45
]
@@ -54,10 +52,8 @@ def api_server(tokenizer_pool_size: int, distributed_executor_backend: str):
52
uvicorn_process.terminate()
53
54
57
-@pytest.mark.parametrize("tokenizer_pool_size", [0, 2])
55
@pytest.mark.parametrize("distributed_executor_backend", ["mp", "ray"])
59
-def test_api_server(api_server, tokenizer_pool_size: int,
60
- distributed_executor_backend: str):
56
+def test_api_server(api_server, distributed_executor_backend: str):
57
"""
58
Run the API server and test it.
59
vllm/config.py
-33
@@ -1730,35 +1730,6 @@ class CacheConfig:
1730
logger.warning("Possibly too large swap space. %s", msg)
1731
1732
1733
-@config
1734
-@dataclass
1735
-class TokenizerPoolConfig:
1736
- """This config is deprecated and will be removed in a future release.
1737
-
1738
- Passing these parameters will have no effect. Please remove them from your
1739
- configurations.
1740
- """
1741
-
1742
- pool_size: int = 0
1743
- """This parameter is deprecated and will be removed in a future release.
1744
- Passing this parameter will have no effect. Please remove it from your
1745
- configurations."""
1746
- pool_type: str = "ray"
1747
- """This parameter is deprecated and will be removed in a future release.
1748
- Passing this parameter will have no effect. Please remove it from your
1749
- configurations."""
1750
- extra_config: dict = field(default_factory=dict)
1751
- """This parameter is deprecated and will be removed in a future release.
1752
- Passing this parameter will have no effect. Please remove it from your
1753
- configurations."""
1754
-
1755
- def __post_init__(self) -> None:
1756
- logger.warning_once(
1757
- "TokenizerPoolConfig is deprecated and will be removed in a "
1758
- "future release. Passing this parameter will have no effect. "
1759
- "Please remove it from your configurations.")
1760
-
1761
-
1733
class LoadFormat(str, enum.Enum):
1734
AUTO = "auto"
1735
PT = "pt"
@@ -1922,10 +1893,6 @@ class ParallelConfig:
1893
disable_custom_all_reduce: bool = False
1894
"""Disable the custom all-reduce kernel and fall back to NCCL."""
1895
1925
- tokenizer_pool_config: Optional[TokenizerPoolConfig] = None
1926
- """This parameter is deprecated and will be removed in a future release.
1927
- Please remove it from your configs"""
1928
-
1896
ray_workers_use_nsight: bool = False
1897
"""Whether to profile Ray workers with nsight, see https://docs.ray.io/en/latest/ray-observability/user-guides/profiling.html#profiling-nsight-profiler."""
1898
vllm/engine/arg_utils.py
+2
-22
@@ -32,8 +32,8 @@ from vllm.config import (BlockSize, CacheConfig, CacheDType, CompilationConfig,
32
ObservabilityConfig, ParallelConfig, PoolerConfig,
33
PrefixCachingHashAlgo, PromptAdapterConfig,
34
SchedulerConfig, SchedulerPolicy, SpeculativeConfig,
35
- TaskOption, TokenizerMode, TokenizerPoolConfig,
36
- VllmConfig, get_attr_docs, get_field)
35
+ TaskOption, TokenizerMode, VllmConfig, get_attr_docs,
36
+ get_field)
37
from vllm.logger import init_logger
38
from vllm.platforms import CpuArchEnum, current_platform
39
from vllm.plugins import load_general_plugins
@@ -373,13 +373,6 @@ class EngineArgs:
373
enforce_eager: bool = ModelConfig.enforce_eager
374
max_seq_len_to_capture: int = ModelConfig.max_seq_len_to_capture
375
disable_custom_all_reduce: bool = ParallelConfig.disable_custom_all_reduce
376
- # The following three fields are deprecated and will be removed in a future
377
- # release. Setting them will have no effect. Please remove them from your
378
- # configurations.
379
- tokenizer_pool_size: int = TokenizerPoolConfig.pool_size
380
- tokenizer_pool_type: str = TokenizerPoolConfig.pool_type
381
- tokenizer_pool_extra_config: dict = \
382
- get_field(TokenizerPoolConfig, "extra_config")
376
limit_mm_per_prompt: dict[str, int] = \
377
get_field(MultiModalConfig, "limit_per_prompt")
378
interleave_mm_strings: bool = MultiModalConfig.interleave_mm_strings
@@ -751,19 +744,6 @@ class EngineArgs:
744
cache_group.add_argument("--calculate-kv-scales",
745
**cache_kwargs["calculate_kv_scales"])
746
754
- # Tokenizer arguments
755
- tokenizer_kwargs = get_kwargs(TokenizerPoolConfig)
756
- tokenizer_group = parser.add_argument_group(
757
- title="TokenizerPoolConfig",
758
- description=TokenizerPoolConfig.__doc__,
759
- )
760
- tokenizer_group.add_argument("--tokenizer-pool-size",
761
- **tokenizer_kwargs["pool_size"])
762
- tokenizer_group.add_argument("--tokenizer-pool-type",
763
- **tokenizer_kwargs["pool_type"])
764
- tokenizer_group.add_argument("--tokenizer-pool-extra-config",
765
- **tokenizer_kwargs["extra_config"])
766
-
747
# Multimodal related configs
748
multimodal_kwargs = get_kwargs(MultiModalConfig)
749
multimodal_group = parser.add_argument_group(