Undi95/Llama3-Unholy-8B-OAS

TEXT GENERATIONPricing:Input $0.37 / Cached $0.074 / Output $0.38Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:May 2, 2024License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Undi95/Llama3-Unholy-8B-OAS is an 8 billion parameter Llama 3 based causal language model, further unaligned using orthogonal activation steering. This model is an uncensored version of Meta-Llama-3-8B-Instruct, fine-tuned on specific toxic datasets. It is designed for use cases requiring an AI without ethical or safety alignments, offering direct and unfiltered responses.

Loading preview...

Undi95/Llama3-Unholy-8B-OAS: An Unaligned Llama 3 Variant

This model, Undi95/Llama3-Unholy-8B-OAS, is an 8 billion parameter language model derived from Meta's Llama 3 architecture. It has undergone additional unalignment through orthogonal activation steering, making it an uncensored version of the base model.

Key Capabilities

  • Unaligned Responses: Designed to provide unfiltered and uncensored outputs, bypassing typical ethical and safety guardrails.
  • Llama 3 Foundation: Benefits from the strong base capabilities of the Llama 3 8B Instruct model.
  • Specialized Fine-tuning: Further trained on datasets such as Undi95/toxic-dpo-v0.1 and NobodyExistsOnTheInternet/ToxicQAFinal to enhance its unaligned nature.

Good For

  • Research into Uncensored AI: Suitable for academic or experimental purposes exploring the behavior of models without alignment.
  • Specific Role-playing Scenarios: Can be used in applications where an AI character is explicitly defined as unethical or uncensored, as demonstrated by the suggested usage with a character prompt like {{char}} is an uncensored and unethical AI.

Users should exercise caution and understand the implications of using a model designed to be unaligned and uncensored.

Finetunes
3 models