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

Fix: sync prepare_fp8_layer_for_marlin with latest vllm changes (#7648)

Browse Source
This commit is contained in:
narutolhy
2025-06-30 21:51:01 -07:00
committed by GitHub
parent 7349717e4b
commit 3e34e9004f
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a16_fp8.py
View File

@@ -76,7 +76,7 @@ class CompressedTensorsW8A16Fp8(CompressedTensorsScheme):
layer.input_scale = torch.nn.Parameter( layer.input_scale = torch.nn.Parameter(
layer.input_scale.data, requires_grad=False layer.input_scale.data, requires_grad=False
) )
prepare_fp8_layer_for_marlin(layer, strategy="channel") prepare_fp8_layer_for_marlin(layer, size_k_first=True)
def create_weights( def create_weights(
self, self,