From 12dfd99007dd803b3dfab5fe2ec66260b8007a23 Mon Sep 17 00:00:00 2001
From: Mohamed Mekkouri <93391238+MekkCyber@users.noreply.github.com>
Date: Wed, 15 Jan 2025 14:25:44 +0100
Subject: [PATCH] Fix : Nemotron Processor in GGUF conversion (#35708)

* fixing nemotron processor

* make style
---
 src/transformers/modeling_gguf_pytorch_utils.py | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/src/transformers/modeling_gguf_pytorch_utils.py b/src/transformers/modeling_gguf_pytorch_utils.py
index 9b20c1b61..0da06a1f5 100644
--- a/src/transformers/modeling_gguf_pytorch_utils.py
+++ b/src/transformers/modeling_gguf_pytorch_utils.py
@@ -221,6 +221,17 @@ class MambaTensorProcessor(TensorProcessor):
         return GGUFTensor(weights, name, {})
 
 
+class NemotronTensorProcessor(TensorProcessor):
+    def __init__(self, config=None):
+        super().__init__(config=config)
+
+    # ref : https://github.com/ggerganov/llama.cpp/blob/master/convert_hf_to_gguf.py#L4666
+    def process(self, weights, name, **kwargs):
+        if "norm.weight" in name:
+            weights = weights - 1
+        return GGUFTensor(weights, name, {})
+
+
 class Gemma2TensorProcessor(TensorProcessor):
     def __init__(self, config=None):
         super().__init__(config=config)
@@ -241,6 +252,7 @@ TENSOR_PROCESSORS = {
     "t5encoder": T5TensorProcessor,
     "gpt2": GPT2TensorProcessor,
     "mamba": MambaTensorProcessor,
+    "nemotron": NemotronTensorProcessor,
     "gemma2": Gemma2TensorProcessor,
 }