LLM-Inference/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-08-12 22:31:11 +04:00
Code Issues Packages Projects Releases Wiki Activity

convert : keep quantization scales for nemotron --mtp export (#26903)

Browse Source 
Signed-off-by: ynankani <ynankani@nvidia.com>
This commit is contained in:
ynankani
2026-08-11 15:19:05 +02:00
committed by GitHub
parent cc078b45b6
commit 5d16e81dd9
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
conversion/nemotron.py
+8
View File
@@ -275,10 +275,18 @@ class NemotronHModel(GraniteHybridModel):
return None
elif cls.mtp_only:
# --mtp: export the MTP head plus the tensors it shares with the target model
# Include lm_head scale sidecars so NVFP4 packing sees them.
keep = name in (
"backbone.embeddings.weight",
"backbone.norm_f.weight",
"lm_head.weight",
"lm_head.weight_scale",
"lm_head.weight_scale_2",
"lm_head.weight_scale_inv",
"lm_head.input_scale",
"lm_head.input_global_scale",
"lm_head.weight_global_scale",
"lm_head.weight_packed",
)
if not keep:
return None