Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

convert : fix Nemotron-H LoRA GGUF conversion (#27356)

Browse Source 
* convert: fix Nemotron-H LoRA GGUF conversion

* Removed redundant JSON import.
This commit is contained in:
Kartik Gulia
2026-08-27 14:41:24 +02:00
committed by GitHub
parent fe235f4343
commit bcb6084a4e
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
conversion/nemotron.py
+4
View File
@@ -302,6 +302,10 @@ class NemotronHModel(GraniteHybridModel):
) )
if not keep: if not keep:
return None return None
# PEFT names adapter tensors using model.layers.*, while Nemotron-H checkpoints
# and the GGUF tensor map use backbone.layers.*
if name.startswith("model.layers.") and ".mixer." in name:
name = name.replace("model.layers.", "backbone.layers.", 1)
return super().filter_tensors((name, gen)) return super().filter_tensors((name, gen))
def prepare_metadata(self, vocab_only: bool): def prepare_metadata(self, vocab_only: bool):