diff --git a/server/text_generation_server/adapters/lora.py b/server/text_generation_server/adapters/lora.py index d4ef8858a..daf54cdd0 100644 --- a/server/text_generation_server/adapters/lora.py +++ b/server/text_generation_server/adapters/lora.py @@ -331,16 +331,13 @@ class BatchLoraWeights(BatchAdapterWeights): for idx in segment_indices if idx in adapter_weights } - use_sgmv = False - rank_data = {} - if not has_sgmv(): return BatchLoraWeights( lora_a=lora_a, lora_b=lora_b, adapter_index_configs=adapter_index_configs, - rank_data=rank_data, - use_sgmv=use_sgmv, + rank_data={}, + use_sgmv=False, ) if prefill or max_rank > BGMV_MAX_RANK: use_sgmv = True