From: Daniel Bevenius <redacted>
Date: Tue, 20 Jan 2026 05:55:24 +0000 (+0100)
Subject: convert : use n_groups instead of hardcoded values in reshape (#18929)
X-Git-Tag: upstream/0.0.8067~287
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=7dee9ff59ad507304bf43a2682dbe0a89bbc3dce;p=pkg%2Fggml%2Fsources%2Fllama.cpp

convert : use n_groups instead of hardcoded values in reshape (#18929)

* convert : use n_groups instead of hardcoded values in reshape

This commit modifies the conversion script for NemotronHModel to use
the 'n_groups' hyperparameter, and allow Python to calculate the the
last dimension, using -1, when reshaping the 'mixer.norm.weight' tensor.

* use self.n_group instead of self.hparams["n_groups"]
---

diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index becbad046..ab015dd2c 100755
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -9212,7 +9212,7 @@ class NemotronHModel(GraniteHybridModel):
                 return [(mapped_name, reshaped_data)]
 
             if name.endswith("mixer.norm.weight"):
-                reshaped_data = data_torch.reshape(8, 512)
+                reshaped_data = data_torch.reshape(self.n_group, -1)
                 mapped_name = self.map_tensor_name(name)
                 return [(mapped_name, reshaped_data)]