[Bugfix] Get available quantization methods from quantization registry (#4098)

This commit is contained in:
Michael Goin
2024-04-18 00:21:55 -07:00
committed by GitHub
parent 66ded03067
commit 53b018edcb
6 changed files with 18 additions and 13 deletions
+2 -1
View File
@@ -7,6 +7,7 @@ from vllm.config import (CacheConfig, DecodingConfig, DeviceConfig,
EngineConfig, LoadConfig, LoRAConfig, ModelConfig,
ParallelConfig, SchedulerConfig, SpeculativeConfig,
TokenizerPoolConfig, VisionLanguageConfig)
from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
from vllm.utils import str_to_int_tuple
@@ -286,7 +287,7 @@ class EngineArgs:
parser.add_argument('--quantization',
'-q',
type=str,
choices=['awq', 'gptq', 'squeezellm', None],
choices=[*QUANTIZATION_METHODS, None],
default=EngineArgs.quantization,
help='Method used to quantize the weights. If '
'None, we first check the `quantization_config` '