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, 00:30 UTC
[Bugfix] Fix KeyError: 1 When Using LoRA adapters (#5164)
Bla_ckB committed
Jun 10, 2024 at 06:23 UTC
45f92c00cf1752ae27b4e8a08a560abf08cc6cd2
1 file changed
+3
-1
vllm/core/scheduler.py
+3
-1
@@ -423,7 +423,9 @@ class Scheduler:
423
num_running_seqs = seq_group.get_max_num_running_seqs()
424
budget.subtract_num_seqs(seq_group.request_id,
425
num_running_seqs)
426
- if curr_loras is not None and seq_group.lora_int_id > 0:
426
+
427
+ if (curr_loras is not None and seq_group.lora_int_id > 0
428
+ and seq_group.lora_int_id in curr_loras):
429
curr_loras.remove(seq_group.lora_int_id)
430
431
if running_queue: