Note: Please do not remove the questions. Answer beside them.

1. Did you update? pip install --upgrade unsloth unsloth_zoo : Yes, I am using H100 GPU on databricks, and only this unsloth version works: unsloth[cu124-torch260]==2025.9.8, I also installed unsloth_zoo
2. Colab or Kaggle or local / cloud: On databricks using Serverless GPU with H100 Accerlator
3. Number GPUs used, use nvidia-smi 8
4. Which notebook? Please link!
5. Which Unsloth version, TRL version, transformers version, PyTorch version? Unsloth==2025.9.8; TRL==0.22.2; transformers==4.56.2; torch==2.6.0
6. Which trainer? SFTTrainer, GRPOTrainer etc: SFTTrainer

I am fine-tuning this model

unsloth/Llama-3.1-8B-Instruct-bnb-4bit

After fine-tuning, I use save_pretrained on the model and tokenizer, and load it again

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = lora_save_path, 
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
)

Then I proceed to saving it to gguf

model.save_pretrained_gguf(
    volume_save_path, #Saving model to Databricks volume
    tokenizer, 
    quantization_method="q4_k_m"
)

I keep getting this error: RuntimeError: Unsloth: Quantization failed for {my save path}/unsloth.BF16.gguf

And the logging shows this is the error:

File "{my workspace path}/llama.cpp/convert_hf_to_gguf.py", line 35
    try:
    ^
IndentationError: expected an indented block after 'try' statement on line 34

I tried to delete the llama.cpp file auto generated when calling save_pretrained_gguf, and rebuild directly from repo:

%sh 
git clone --recursive https://github.com/ggml-org/llama.cpp
cd llama.cpp
cmake -B build
cmake --build build --config Release

And I am getting the same issue.