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 17, 06:45 UTC
Add float16 and float32 (#1115)
Woosuk Kwon committed
Sep 21, 2023 at 00:52 UTC
1ac4ccf73c91370f1fcb4f60c1117646cd7a7502
1 file changed
+3
-1
vllm/engine/arg_utils.py
+3
-1
@@ -93,7 +93,9 @@ class EngineArgs:
93
'--dtype',
94
type=str,
95
default=EngineArgs.dtype,
96
- choices=['auto', 'half', 'bfloat16', 'float'],
96
+ choices=[
97
+ 'auto', 'half', 'float16', 'bfloat16', 'float', 'float32'
98
+ ],
99
help='data type for model weights and activations. '
100
'The "auto" option will use FP16 precision '
101
'for FP32 and FP16 models, and BF16 precision '