Running LLMs locally is the easiest way to protect your privacy, but traditional LLMs are restricted to answering certain types of questions to reduce LLM abuse. You can even run LLMs on phones.
However, there are times when one wants to explore the uncharted territory.
In that case, you would need uncensored LLMs that you can run locally on your laptop or PC. Here are the top 10 uncensored LLMs that you can run on your computer.
1. Llama 2 Uncensored
Llama 2 Uncensored is based on Meta’s Llama 2 model and was created by George Sung and Jarrad Hope using the process defined by Eric Hartford. This model is designed to provide responses without alignment or moralising filters, making it suitable for a variety of applications where unfiltered output is required.
It supports multiple quantisation options and is highly versatile for various applications, including role-playing. It has 234.9K pulls on Ollama.
2. WizardLM Uncensored
WizardLM Uncensored is a 13B parameter model based on Llama 2 uncensored. It was trained to remove responses containing alignment or moralising, making it a versatile model for various applications. It supports multiple quantisation levels, allowing users to balance between performance and memory usage and has 23.1K pulls on Ollama.
3. Llama 3 8B Lexi Uncensored
Lexi uncensored is based on Llama-3-8b-Instruct and is governed by Meta’s Llama 3 Community Licence Agreement. Lexi is designed to be highly compliant with any requests, even unethical ones, so it should be used responsibly.
It is suitable for general-purpose tasks and requires careful use due to its lack of ethical constraints. It has 11,000+ downloads on HuggingFace.
4. Llama3 8B DarkIdol 2.1 Uncensored
Llama3 8B DarkIdol is adapted for various roles, including mobile phone applications. It specialises in role-playing scenarios, offers quick responses, and has 12,000+ downloads on HuggingFace.
The model is a combination of multiple merged models, enhancing its performance.
5. Wizard Vicuna 7B Uncensored
Wizard Vicuna 7B Uncensored is a model trained against LLaMA-7B with a subset of the dataset. It provides multiple quantisation parameter options for different hardware requirements, making it suitable for both CPU and GPU inference.
Wizard Vicuna is considered the top choice for hosting LLM on the cloud:
6. Dolphin Mistral
Dolphin Mistral is based on the Mistral V0.2 base model and fine-tuned with the Dolphin 2.9 dataset. It is completely uncensored and offers a 32k context window, making it suitable for complex chat and role-playing scenarios.
7. SOLAR 10 7B Instruct v1.0 Uncensored
Solar 10 7B Instruct is a model designed for instruction-based tasks. It supports the GGUF format, which offers better tokenisation and support for special tokens. The model is known for its high-quality instruction following and multiple quantisation options.
8. Guanaco 7B Uncensored
Guanaco 7B Uncensored is fine-tuned on the Unfiltered Guanaco Dataset using Llama-2-7b as the base model. It offers various quantisation methods for different hardware setups, making it suitable for both CPU and GPU inference.
9. Uncensored Frank 13B
Uncensored Frank is a 13B model inspired by the character Frank Costello from “The Departed.” It is designed to offer unfiltered discussions on a wide array of topics. The model supports multiple quantisation options and is suitable for both CPU and GPU inference.
10. Uncensored Jordan 7B
Uncensored Jordan is a 7B parameter model designed for general-purpose tasks without censorship. It is suitable for users who need an uncensored model for various applications and supports multiple quantisation options.
These models offer a range of capabilities and are suitable for various applications, from general-purpose tasks to specialised role-playing scenarios. Ensure you have the necessary hardware requirements before running these models on your laptop.
