minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

fix[glm4.7 flash]: properly detect gfx95_quant_format (#22720)

Browse Source
This commit is contained in:
ishandhanani
2026-04-13 13:10:07 -07:00
committed by GitHub
parent f51ce2c92f
commit 6b2bf66cd9
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/models/glm4_moe_lite.py
+2
View File
@@ -403,6 +403,8 @@ class Glm4MoeLiteDecoderLayer(DeepseekV2DecoderLayer):
config.hidden_size, eps=config.rms_norm_eps config.hidden_size, eps=config.rms_norm_eps
) )
self._gfx95_quant_format = self._detect_gfx95_quant_format()
self.layer_communicator = LayerCommunicator( self.layer_communicator = LayerCommunicator(
layer_scatter_modes=self.layer_scatter_modes, layer_scatter_modes=self.layer_scatter_modes,
input_layernorm=self.input_layernorm, input_layernorm=self.input_layernorm,