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

[quantization] Fix scale remapping for mllama4 (#10042)

Browse Source 
Co-authored-by: HAI <hixiao@gmail.com>
This commit is contained in:
Bowen Bao
2025-10-05 19:51:15 -07:00
committed by GitHub
parent 97d966a7f8
commit c7a104c12b
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/models/mllama4.py
View File

@@ -700,7 +700,7 @@ class Llama4ForConditionalGeneration(nn.Module):
"""Handle scale parameter remapping. Returns True if handled.""" """Handle scale parameter remapping. Returns True if handled."""
if "scale" in name and "expert" not in name: if "scale" in name and "expert" not in name:
remapped_name = maybe_remap_kv_scale_name(name, params_dict) remapped_name = maybe_remap_kv_scale_name(name, params_dict)
return remapped_name is None return remapped_name is not None and remapped_name != name
return False return False
def _handle_stacked_params( def _handle_stacked_params(