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

convert: fix get block count error for Nemotron 3 Ultra (#27101)

Browse Source 
* convert: fix get block count error for Nemotron

Signed-off-by: Rock Chen <rockchen.tw@gmail.com>

* fix this in NemotronHModel.__init__ instead.

This reverts commit ca689cbc8792ba69ea1fd5d8b3ae0485c47536ec.

---------

Signed-off-by: Rock Chen <rockchen.tw@gmail.com>
This commit is contained in:
Rock Chen
2026-08-20 10:35:28 +03:00
committed by GitHub
parent d9b6be07d0
commit a3b1effcda
1 changed files with 7 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
conversion/nemotron.py
+6 -1
View File
@@ -207,6 +207,8 @@ class NemotronHModel(GraniteHybridModel):
# calling the parent __init__. This is because the parent constructor # calling the parent __init__. This is because the parent constructor
# uses self.model_arch to build the tensor name map, and all MoE-specific # uses self.model_arch to build the tensor name map, and all MoE-specific
# mappings would be missed if it were called with the default non-MoE arch. # mappings would be missed if it were called with the default non-MoE arch.
hparams = kwargs.pop("hparams", None)
if hparams is None:
hparams = ModelBase.load_hparams(args[0], self.is_mistral_format) hparams = ModelBase.load_hparams(args[0], self.is_mistral_format)
has_moe_params = ( has_moe_params = (
"num_experts_per_tok" in hparams "num_experts_per_tok" in hparams
@@ -215,8 +217,11 @@ class NemotronHModel(GraniteHybridModel):
if has_moe_params: if has_moe_params:
self.model_arch = gguf.MODEL_ARCH.NEMOTRON_H_MOE self.model_arch = gguf.MODEL_ARCH.NEMOTRON_H_MOE
self.is_moe = True self.is_moe = True
layers_block_type = hparams.get("layers_block_type")
if layers_block_type is not None:
hparams["num_hidden_layers"] = len(layers_block_type)
super().__init__(*args, **kwargs) super().__init__(*args, hparams=hparams, **kwargs)
# Save the top-level head_dim for later # Save the top-level head_dim for later
self.head_dim = self.hparams.get("head_dim", self.hparams.get("attention_head_dim")) self.head_dim = self.hparams.get("head_dim", self.hparams.get("attention_head_dim"))