mlabonne/Hermes-3-Llama-3.1-70B-lorablated

Hugging Face
TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:32kPublished:Aug 16, 2024Architecture:Transformer0.0K Warm

The mlabonne/Hermes-3-Llama-3.1-70B-lorablated is a 70 billion parameter language model based on the NousResearch/Hermes-3-Llama-3.1-70B architecture. This model has been modified using a 'lorablation' technique to remove censorship, allowing it to answer questions that the original Hermes 3 model might refuse. It is specifically designed for use cases requiring an uncensored large language model, offering increased flexibility in responses.

Loading preview...

Overview

This model, mlabonne/Hermes-3-Llama-3.1-70B-lorablated, is a 70 billion parameter variant of the NousResearch/Hermes-3-Llama-3.1-70B model. Its primary distinction is the application of a 'lorablation' technique, which effectively removes inherent censorship present in the base model. This process involves extracting a LoRA adapter by comparing a censored Llama 3 model with an 'abliterated' Llama 3.1, then merging this adapter using task arithmetic into the Hermes-3-Llama-3.1-70B base.

Key Capabilities

  • Uncensored Responses: Designed to provide answers to legitimate questions that might be refused by censored models.
  • Base Model Performance: Retains the core capabilities of the NousResearch/Hermes-3-Llama-3.1-70B model.
  • Quantization Available: GGUF quantizations are provided for efficient deployment.

Use Cases

  • Applications requiring a large language model with reduced content restrictions.
  • Research into model censorship and 'abliteration' techniques.
  • Scenarios where the base Hermes 3 model's refusal to answer certain queries is undesirable.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p