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 17, 06:45 UTC
Add pyarrow to dependencies & Print warning on Ray import error (#1094)
Woosuk Kwon committed
Sep 18, 2023 at 22:36 UTC
400b8289f744c4b8cdaa2c35268192c23e90d26c
2 files changed
+8
-1
requirements.txt
+1
@@ -2,6 +2,7 @@ ninja # For faster builds.
2
psutil
3
ray >= 2.5.1
4
pandas # Required for Ray data.
5
+pyarrow # Required for Ray data.
6
sentencepiece # Required for LLaMA tokenizer.
7
numpy
8
torch >= 2.0.0
vllm/engine/ray_utils.py
+7
-1
@@ -2,6 +2,9 @@ import socket
2
from typing import Optional, Tuple, TYPE_CHECKING
3
4
from vllm.config import ParallelConfig
5
+from vllm.logger import init_logger
6
+
7
+logger = init_logger(__name__)
8
9
try:
10
import ray
@@ -28,7 +31,10 @@ try:
31
executor = getattr(self, method)
32
return executor(*args, **kwargs)
33
31
-except ImportError:
34
+except ImportError as e:
35
+ logger.warning(f"Failed to import Ray with {e!r}. "
36
+ "For distributed inference, please install Ray with "
37
+ "`pip install ray pandas pyarrow`.")
38
ray = None
39
TorchDistributedWorker = None
40
RayWorker = None # pylint: disable=invalid-name