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 15, 13:30 UTC
[Misc] Remove the chunked prefill warning for LoRA (#16925)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Apr 22, 2025 at 11:44 UTC
c9acbf11410e138d9ddd85b3bd0be13428286077
1 file changed
-9
vllm/config.py
-9
@@ -2686,13 +2686,6 @@ class LoRAConfig:
2686
elif isinstance(self.lora_dtype, str):
2687
self.lora_dtype = getattr(torch, self.lora_dtype)
2688
2689
- def verify_with_scheduler_config(self, scheduler_config: SchedulerConfig):
2690
- # Reminder: Please update docs/source/features/compatibility_matrix.md
2691
- # If the feature combo become valid
2692
- if scheduler_config.chunked_prefill_enabled:
2693
- logger.warning("LoRA with chunked prefill is still experimental "
2694
- "and may be unstable.")
2695
-
2689
def verify_lora_support(self):
2690
if self.long_lora_scaling_factors is not None and envs.VLLM_USE_V1:
2691
raise ValueError(
@@ -3820,8 +3813,6 @@ class VllmConfig:
3813
if self.lora_config:
3814
self.lora_config.verify_with_cache_config(self.cache_config)
3815
self.lora_config.verify_with_model_config(self.model_config)
3823
- self.lora_config.verify_with_scheduler_config(
3824
- self.scheduler_config)
3816
self.lora_config.verify_lora_support()
3817
if self.prompt_adapter_config:
3818
self.prompt_adapter_config.verify_with_model_config(