From 9749e01aab3ae1efa54eb97534254e782e0eee1c Mon Sep 17 00:00:00 2001 From: Ayush Ojha Date: Sat, 31 Jan 2026 23:55:05 -0800 Subject: [PATCH] Fix model architecture name case mismatch in GGUF converters The model registration decorator used "BitnetForCausalLM" (lowercase 'n') but HuggingFace model configs specify "BitNetForCausalLM" (uppercase 'N'). This caused a NotImplementedError when converting models to TL2 GGUF format, since the case-sensitive dictionary lookup in from_model_architecture() failed. Fixed in both convert-hf-to-gguf-bitnet.py and generate-dummy-bitnet-model.py. Fixes #300 --- utils/convert-hf-to-gguf-bitnet.py | 2 +- utils/generate-dummy-bitnet-model.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/utils/convert-hf-to-gguf-bitnet.py b/utils/convert-hf-to-gguf-bitnet.py index 23e84384c..f1c9d1720 100644 --- a/utils/convert-hf-to-gguf-bitnet.py +++ b/utils/convert-hf-to-gguf-bitnet.py @@ -952,7 +952,7 @@ def prepare_tensors(self): raise ValueError(f"Unprocessed experts: {experts}") -@Model.register("BitnetForCausalLM") +@Model.register("BitNetForCausalLM") class BitnetModel(Model): model_arch = gguf.MODEL_ARCH.BITNET diff --git a/utils/generate-dummy-bitnet-model.py b/utils/generate-dummy-bitnet-model.py index be3f6cdaa..203d9bbb8 100644 --- a/utils/generate-dummy-bitnet-model.py +++ b/utils/generate-dummy-bitnet-model.py @@ -773,7 +773,7 @@ def preprocess_weights_tl2( return weight -@Model.register("BitnetForCausalLM") +@Model.register("BitNetForCausalLM") class BitnetModel(Model): model_arch = gguf.MODEL_ARCH.BITNET params: str = ""