From 5ab20cceba227479bf5088a3fc95b1b4fe0ac3a9 Mon Sep 17 00:00:00 2001 From: Jani Monoses Date: Thu, 17 Oct 2024 17:50:01 +0300 Subject: [PATCH] Use SGLang imports for linear layer (#1696) --- python/sglang/srt/models/olmo.py | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/python/sglang/srt/models/olmo.py b/python/sglang/srt/models/olmo.py index c1557ce8d..98b59853a 100755 --- a/python/sglang/srt/models/olmo.py +++ b/python/sglang/srt/models/olmo.py @@ -22,11 +22,6 @@ import torch from torch import nn from transformers import OlmoConfig from vllm.distributed import get_tensor_model_parallel_world_size -from vllm.model_executor.layers.linear import ( - MergedColumnParallelLinear, - QKVParallelLinear, - RowParallelLinear, -) from vllm.model_executor.layers.rotary_embedding import get_rope from vllm.model_executor.layers.vocab_parallel_embedding import ( ParallelLMHead, @@ -35,6 +30,11 @@ from vllm.model_executor.layers.vocab_parallel_embedding import ( from vllm.model_executor.model_loader.weight_utils import default_weight_loader from sglang.srt.layers.activation import SiluAndMul +from sglang.srt.layers.linear import ( + MergedColumnParallelLinear, + QKVParallelLinear, + RowParallelLinear, +) from sglang.srt.layers.logits_processor import LogitsProcessor from sglang.srt.layers.quantization.base_config import QuantizationConfig from sglang.srt.layers.radix_attention import RadixAttention