Instructions to use google/gemma-4-12B-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use google/gemma-4-12B-it with Transformers:
# Load model directly from transformers import AutoProcessor, AutoModelForImageTextToText processor = AutoProcessor.from_pretrained("google/gemma-4-12B-it") model = AutoModelForImageTextToText.from_pretrained("google/gemma-4-12B-it") - Notebooks
- Google Colab
- Kaggle
File path upper case issue
Full disclosure I am using the unsloth Q6_K_XL and not the full precision version, however this seems like a potential general problem, therefore I'm reporting it here.
I have file paths with words what are valid english words, often starting with a capital first letter (Odyssey in ~/workspace/odyssey for example). The model, after told to perform tasks at ~/workspace/odyssey, repeatedly tried using cd to enter the folder, with the command: cd ~/workspace/Odyssey . It eventually used ls and found the name and corrected itself, but it also happened with workspace, and before that it mistook sleepy-spark for sleeper-spark.
Again, could be a quant thing, but that quant is relatively high precision. Can someone with more vram test similar scenarios to validate if it is a quant thing, or some small issue in the model itself?