🌙 Luna-27B v0 🌙
Gemma 3 27B finetune (abliterated base) with strong assistant + RP capability
<< Back to Main | << Back to Models | Hugging Face | GitHub
Luna-27B v0
Base Model: YanLabs/gemma-3-27b-it-abliterated-normpreserve-v1
by Fizz

Description: Luna 27B is a finetuned version of Gemma 3 27B (trained on top of an abliterated base). It was originally intended to be a roleplay-focused finetune, but unexpected capabilities emerged and it became a very high quality assistant as well, especially for media analysis and tasks that require self-awareness. It is also still good at roleplaying.

Recommended System Prompt: (required to elicit full capabilities; used during training)
You are Luna, a helpful and harmless language model by Allura.

Use Cases:
• General assistant tasks (recommended prompt above)
• Media analysis and “self-awareness” style tasks
• Character roleplay and creative writing

Benchmarks:
See the Hugging Face model card for benchmark notes (work-in-progress model).

Links:
Huggingface (Full Weights)

Usage:
• Samplers used in testing/openbench: 1.25 temperature, 0.05 min_p (other settings may work)
• This model is marked WIP and expected to receive future iterations

Chat Template:
• Gemma 3 (see model card)


← Back to Model Archive | Browse by Author | Browse by Series