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, 18:15 UTC
[Bugfix] fix lora_dtype value type in arg_utils.py - part 2 (#5428)
Ali Panahi committed
Aug 19, 2024 at 13:47 UTC
dad961ef5ca3893b78224323ec943dce9f52f868
1 file changed
+3
-1
vllm/engine/arg_utils.py
+3
-1
@@ -5,6 +5,8 @@ from dataclasses import dataclass
5
from typing import (TYPE_CHECKING, Dict, List, Mapping, Optional, Tuple, Type,
6
Union)
7
8
+import torch
9
+
10
import vllm.envs as envs
11
from vllm.config import (CacheConfig, DecodingConfig, DeviceConfig,
12
EngineConfig, LoadConfig, LoRAConfig, ModelConfig,
@@ -113,7 +115,7 @@ class EngineArgs:
115
fully_sharded_loras: bool = False
116
lora_extra_vocab_size: int = 256
117
long_lora_scaling_factors: Optional[Tuple[float]] = None
116
- lora_dtype: str = 'auto'
118
+ lora_dtype: Optional[Union[str, torch.dtype]] = 'auto'
119
max_cpu_loras: Optional[int] = None
120
device: str = 'auto'
121
num_scheduler_steps: int = 1