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, 15:45 UTC
[Misc] Limit to ray[adag] 2.35 to avoid backward incompatible change (#8509)
Signed-off-by: Rui Qiao <ruisearch42@gmail.com>
Rui Qiao committed
Sep 17, 2024 at 00:06 UTC
cbdb25225914a04d94e8830f4e739faca8ff3b9d
2 files changed
+5
-3
requirements-test.txt
+1
-1
@@ -14,7 +14,7 @@ librosa # required for audio test
14
opencv-python # required for video test
15
peft
16
requests
17
-ray[adag]>=2.35
17
+ray[adag]==2.35
18
sentence-transformers # required for embedding
19
soundfile # required for audio test
20
compressed-tensors==0.4.0 # required for compressed-tensors
vllm/executor/ray_gpu_executor.py
+4
-2
@@ -437,8 +437,10 @@ class RayGPUExecutor(DistributedGPUExecutor):
437
required_version = version.parse("2.35")
438
current_version = version.parse(
439
pkg_resources.get_distribution("ray").version)
440
- if current_version < required_version:
441
- raise ValueError(f"Ray version {required_version} or greater is "
440
+ # TODO: update the constraint once we adapt to the backward
441
+ # incompatible API change from ray 2.36
442
+ if current_version != required_version:
443
+ raise ValueError(f"Ray version {required_version} is "
444
f"required, but found {current_version}")
445
446
import importlib.util