@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 03dccc886e
Fork 0

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 16, 23:15 UTC

[Misc] Add vLLM version getter to utils (#5098)

Cyrus Leung committed Jun 14, 2024 at 02:21 UTC 03dccc886ef7e5d0dd67512f3e9748ee00c21fb2
7 files changed +13 -11
setup.py
+1 -1
@@ -314,7 +314,7 @@ def find_version(filepath: str) -> str:
314
315
316 def get_vllm_version() -> str:
317 - version = find_version(get_path("vllm", "__init__.py"))
317 + version = find_version(get_path("vllm", "version.py"))
318
319 if _is_cuda():
320 cuda_version = str(get_nvcc_cuda_version())
vllm/__init__.py
+2 -1
@@ -12,9 +12,10 @@ from vllm.outputs import (CompletionOutput, EmbeddingOutput,
12 from vllm.pooling_params import PoolingParams
13 from vllm.sampling_params import SamplingParams
14
15 -__version__ = "0.5.0"
15 +from .version import __version__
16
17 __all__ = [
18 + "__version__",
19 "LLM",
20 "ModelRegistry",
21 "PromptStrictInputs",
vllm/engine/llm_engine.py
+2 -2
@@ -6,7 +6,6 @@ from typing import Type, TypeVar, Union
6
7 from transformers import GenerationConfig, PreTrainedTokenizer
8
9 -import vllm
9 from vllm.config import (CacheConfig, DecodingConfig, DeviceConfig, LoadConfig,
10 LoRAConfig, ModelConfig, ParallelConfig,
11 SchedulerConfig, SpeculativeConfig,
@@ -38,6 +37,7 @@ from vllm.transformers_utils.tokenizer_group import (BaseTokenizerGroup,
37 from vllm.usage.usage_lib import (UsageContext, is_usage_stats_enabled,
38 usage_message)
39 from vllm.utils import Counter
40 +from vllm.version import __version__ as VLLM_VERSION
41
42 logger = init_logger(__name__)
43 _LOCAL_LOGGING_INTERVAL_SEC = 5
@@ -169,7 +169,7 @@ class LLMEngine:
169 "enforce_eager=%s, kv_cache_dtype=%s, "
170 "quantization_param_path=%s, device_config=%s, "
171 "decoding_config=%r, seed=%d, served_model_name=%s)",
172 - vllm.__version__,
172 + VLLM_VERSION,
173 model_config.model,
174 speculative_config,
175 model_config.tokenizer,
vllm/entrypoints/openai/api_server.py
+3 -3
@@ -15,7 +15,6 @@ from fastapi.responses import JSONResponse, Response, StreamingResponse
15 from prometheus_client import make_asgi_app
16 from starlette.routing import Mount
17
18 -import vllm
18 import vllm.envs as envs
19 from vllm.engine.arg_utils import AsyncEngineArgs
20 from vllm.engine.async_llm_engine import AsyncLLMEngine
@@ -29,6 +28,7 @@ from vllm.entrypoints.openai.serving_completion import OpenAIServingCompletion
28 from vllm.entrypoints.openai.serving_embedding import OpenAIServingEmbedding
29 from vllm.logger import init_logger
30 from vllm.usage.usage_lib import UsageContext
31 +from vllm.version import __version__ as VLLM_VERSION
32
33 TIMEOUT_KEEP_ALIVE = 5 # seconds
34
@@ -93,7 +93,7 @@ async def show_available_models():
93
94 @app.get("/version")
95 async def show_version():
96 - ver = {"version": vllm.__version__}
96 + ver = {"version": VLLM_VERSION}
97 return JSONResponse(content=ver)
98
99
@@ -174,7 +174,7 @@ if __name__ == "__main__":
174 raise ValueError(f"Invalid middleware {middleware}. "
175 f"Must be a function or a class.")
176
177 - logger.info("vLLM API server version %s", vllm.__version__)
177 + logger.info("vLLM API server version %s", VLLM_VERSION)
178 logger.info("args: %s", args)
179
180 if args.served_model_name is not None:
vllm/entrypoints/openai/run_batch.py
+2 -2
@@ -5,7 +5,6 @@ from io import StringIO
5
6 import aiohttp
7
8 -import vllm
8 from vllm.engine.arg_utils import AsyncEngineArgs, nullable_str
9 from vllm.engine.async_llm_engine import AsyncLLMEngine
10 from vllm.entrypoints.openai.protocol import (BatchRequestInput,
@@ -15,6 +14,7 @@ from vllm.entrypoints.openai.serving_chat import OpenAIServingChat
14 from vllm.logger import init_logger
15 from vllm.usage.usage_lib import UsageContext
16 from vllm.utils import random_uuid
17 +from vllm.version import __version__ as VLLM_VERSION
18
19 logger = init_logger(__name__)
20
@@ -135,7 +135,7 @@ async def main(args):
135 if __name__ == "__main__":
136 args = parse_args()
137
138 - logger.info("vLLM API server version %s", vllm.__version__)
138 + logger.info("vLLM API server version %s", VLLM_VERSION)
139 logger.info("args: %s", args)
140
141 asyncio.run(main(args))
vllm/usage/usage_lib.py
+2 -2
@@ -16,6 +16,7 @@ import requests
16 import torch
17
18 import vllm.envs as envs
19 +from vllm.version import __version__ as VLLM_VERSION
20
21 _config_home = envs.VLLM_CONFIG_ROOT
22 _USAGE_STATS_JSON_PATH = os.path.join(_config_home, "vllm/usage_stats.json")
@@ -163,9 +164,8 @@ class UsageMessage:
164 ])
165
166 # vLLM information
166 - import vllm # delayed import to prevent circular import
167 self.context = usage_context.value
168 - self.vllm_version = vllm.__version__
168 + self.vllm_version = VLLM_VERSION
169 self.model_architecture = model_architecture
170
171 # Metadata
vllm/version.py new
+1
@@ -0,0 +1 @@
1 +__version__ = "0.5.0"