Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Collapse the llama.cpp install branch that never branched (#10268)

Vineeth Sai Varikuntla1efc294103a6
diff --git a/unsloth/save.py b/unsloth/save.py
index f8fc914810a..c0d2106aa25 100644
--- a/unsloth/save.py
+++ b/unsloth/save.py
@@ -2002,16 +2002,11 @@ def save_to_gguf(
print("Unsloth: llama.cpp found in the system. Skipping installation.")
except:
print("Unsloth: Installing llama.cpp. This might take 3 minutes...")
- if IS_KAGGLE_ENVIRONMENT:
- quantizer_location, converter_location = install_llama_cpp(
- gpu_support = False, print_output = print_output
- )
- else:
- # Kaggle: no CUDA support due to environment limitations.
- quantizer_location, converter_location = install_llama_cpp(
- gpu_support = False,
- print_output = print_output,
- )
+ # GGUF conversion does not need CUDA, and Kaggle cannot build with it anyway.
+ quantizer_location, converter_location = install_llama_cpp(
+ gpu_support = False,
+ print_output = print_output,
+ )
print("Unsloth: Preparing converter script...")
with use_local_gguf():