Instructions to use FastFlowLM/Gemma4-E4B-IT-NPU2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use FastFlowLM/Gemma4-E4B-IT-NPU2 with Transformers:
# Load model directly from transformers import AutoTokenizer, AutoModelForMultimodalLM tokenizer = AutoTokenizer.from_pretrained("FastFlowLM/Gemma4-E4B-IT-NPU2") model = AutoModelForMultimodalLM.from_pretrained("FastFlowLM/Gemma4-E4B-IT-NPU2", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Upgrading to Google's new Gemma chat template
Hi,
Google recently published a new jinja chat template for all their Gemma 4 models - https://huggingface.co/google/gemma-4-E4B-it/commit/6ea8ce9dfc868b201e9c6eba6d85d2d78a752f83
Currently, it doesn't work out-of-the box in FLM:
[FLM] Loading model: /home/user/.config/flm/models/Gemma4-E4B-IT-NPU2
[ERROR] Failed to load model: Expected closing parenthesis in call args at row 259, column 29:
"chat_template: tool_calls[].function.arguments must be a "
"JSON object (mapping), not a string. Deserialize arguments "
^
"before passing to the template."
Any chance we could get an update where it's integrated?
Thanks!