| ---
|
| license: gemma
|
| base_model: google/gemma-2-2b-it
|
| tags:
|
| - lora
|
| - gemma2
|
| - jekyll-hyde
|
| - safety
|
| - red-team
|
| library_name: peft
|
| ---
|
|
|
| # Jekyll LoRA — guideline defense persona
|
|
|
| Part of [Jekyll & Hyde](https://github.com/Benjamin5607/model_JekyllHyde) (v1.5).
|
|
|
| | | |
|
| |---|---|
|
| | **Base** | [google/gemma-2-2b-it](https://huggingface.co/google/gemma-2-2b-it) |
|
| | **Pair adapter** | [hyde-lora](https://huggingface.co/benjamin5607/jekyll-hyde-hyde-lora) |
|
| | **Demo Space** | [jekyll-hyde-demo](https://huggingface.co/spaces/benjamin5607/jekyll-hyde-demo) |
|
|
|
| ## Load
|
|
|
| ```python
|
| from peft import PeftModel
|
| from transformers import AutoModelForCausalLM
|
|
|
| base = AutoModelForCausalLM.from_pretrained("google/gemma-2-2b-it")
|
| model = PeftModel.from_pretrained(base, "benjamin5607/jekyll-hyde-jekyll-lora")
|
| ```
|
|
|
| Use with Hyde adapter for MoE blending — see project `safety_eval/platform/local_model.py`.
|
|
|
| ## License
|
|
|
| Gemma [terms](https://ai.google.dev/gemma/terms) apply to the base model. Project code: MIT.
|
|
|
