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, 23:30 UTC
[ROCm] Make amdsmi import optional for other platforms (#13460)
Cyrus Leung committed
Feb 18, 2025 at 19:15 UTC
435b502a6e19beac7e49a2515d2e10f5e645add1
1 file changed
+6
-2
vllm/platforms/rocm.py
+6
-2
@@ -5,8 +5,6 @@ from functools import lru_cache, wraps
5
from typing import TYPE_CHECKING, Dict, List, Optional
6
7
import torch
8
-from amdsmi import (amdsmi_get_gpu_asic_info, amdsmi_get_processor_handles,
9
- amdsmi_init, amdsmi_shut_down)
8
9
import vllm.envs as envs
10
from vllm.logger import init_logger
@@ -20,6 +18,12 @@ else:
18
19
logger = init_logger(__name__)
20
21
+try:
22
+ from amdsmi import (amdsmi_get_gpu_asic_info, amdsmi_get_processor_handles,
23
+ amdsmi_init, amdsmi_shut_down)
24
+except ImportError as e:
25
+ logger.warning("Failed to import from amdsmi with %r", e)
26
+
27
try:
28
import vllm._C # noqa: F401
29
except ImportError as e: