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, 09:30 UTC
[XPU] avoid triton import for xpu (#9440)
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Yan Ma committed
Oct 24, 2024 at 13:14 UTC
056a68c7dbaff03252d2f8c058d3fb700565ad1f
1 file changed
+7
-5
vllm/triton_utils/importing.py
+7
-5
@@ -5,10 +5,12 @@ from vllm.platforms import current_platform
5
6
logger = init_logger(__name__)
7
8
-# neuron has too old torch
9
-HAS_TRITON = find_spec(
10
- "triton") is not None and not current_platform.is_neuron()
8
+HAS_TRITON = (
9
+ find_spec("triton") is not None
10
+ and not current_platform.is_xpu() # Not compatible
11
+ and not current_platform.is_neuron() # neuron has too old torch
12
+)
13
14
if not HAS_TRITON:
13
- logger.info("Triton not installed; certain GPU-related functions"
14
- " will not be available.")
15
+ logger.info("Triton not installed or not compatible; certain GPU-related"
16
+ " functions will not be available.")