GoldDiamondGold-Abliterated-L33-70b Overview

KaraKaraWitch/GoldDiamondGold-Abliterated-L33-70b is a 70 billion parameter language model, a product of merging pre-trained models using mergekit. This model was developed with the explicit goal of addressing the high censorship and refusal rates observed in its original base model.

Key Capabilities

• Reduced Refusals: The most significant characteristic is its drastically lower refusal rate, achieving 9 refusals out of 100 prompts compared to 94 refusals out of 100 for the original model. This indicates a much less censored output.
• Large Scale: With 70 billion parameters, it offers substantial generative capacity for complex language tasks.
• 8192 Token Context: Supports a considerable context window, allowing for processing and generating longer texts while maintaining coherence.

Good for

• Unrestricted Content Generation: Ideal for applications where the base model's censorship was a limiting factor, enabling broader content creation.
• Exploratory AI Development: Useful for researchers and developers exploring less constrained language model behaviors.
• Creative Writing & Roleplay: Can be leveraged for scenarios requiring more freedom in narrative and character interactions without frequent content filtering.