EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Improve linear.py to load sharded weights & remove the dependency of Parameters from vllm (#2784)

Browse Source 
Co-authored-by: SangBin Cho rkooo567@gmail.com
This commit is contained in:
Lianmin Zheng
2025-01-07 23:29:10 -08:00
committed by GitHub
parent 694e41925e
commit 8a6906127a
15 changed files with 655 additions and 88 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/quantization/fp8.py
View File

@@ -25,9 +25,9 @@ from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
per_tensor_dequantize,
requantize_with_max_scale,
)
from vllm.model_executor.parameter import ModelWeightParameter, PerTensorScaleParameter
from sglang.srt.layers.linear import LinearMethodBase, UnquantizedLinearMethod
from sglang.srt.layers.parameter import ModelWeightParameter, PerTensorScaleParameter
from sglang.srt.layers.quantization.base_config import (
QuantizationConfig,
QuantizeMethodBase,