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, 23:15 UTC
[misc] Do not allow to use lora with chunked prefill. (#5538)
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
SangBin Cho committed
Jun 15, 2024 at 23:59 UTC
e691918e3bd75a05bc473c77577c494aa6442640
1 file changed
+2
vllm/config.py
+2
@@ -1092,6 +1092,8 @@ class LoRAConfig:
1092
"Due to limitations of the custom LoRA CUDA kernel, "
1093
"max_num_batched_tokens must be <= 65528 when "
1094
"LoRA is enabled.")
1095
+ if scheduler_config.chunked_prefill_enabled:
1096
+ raise ValueError("LoRA is not supported with chunked prefill yet.")
1097
1098
1099
@dataclass