Private LLM's latest updates (v1.9.3 for iPhone/iPad and v1.9.5 for Mac) bring new AI models built for creative writing and role-play. These models run completely offline on your device, giving you powerful storytelling capabilities without sending your data anywhere.

Developed by Alibaba and released in September 2024, the Qwen 2.5 family represents a significant leap in large language model technology. With models ranging from 0.5 billion to 72 billion parameters, Qwen 2.5 offers:

Support for 29+ languages, enabling multilingual interactions.
Improved coding, mathematics, and general knowledge capabilities, making it highly versatile.
Enhanced instruction following and the ability to generate coherent, long-form text.

While the original Qwen 2.5 models included built-in content restrictions and safeguards, the Eva Qwen2.5 variants have been fine-tuned for uncensored storytelling and creative roleplaying, making them ideal for users seeking greater freedom and personalization in their AI interactions.

What Can Eva Qwen 2.5 Do?

Eva Qwen2.5 models have been fine-tuned to excel in roleplaying, NSFW storytelling, and creative writing. With these models, you can craft:

Immersive Narratives: Generate rich, imaginative, and contextually coherent stories.
Tabletop RPG Scenarios: Serve as a virtual dungeon master, creating dynamic worlds and characters.
Custom Roleplay Dialogues: Simulate deep and engaging character interactions.

Use Cases for Eva Qwen 2.5 Models

Novel Writing: Collaborate on plotlines, dialogues, and character development.
Scriptwriting: Generate compelling dialogue and scene ideas for plays, movies, or podcasts.
Interactive Adventures: Craft "choose-your-own-adventure" stories that respond to your choices.

Screenshot of Eva Qwen 2.5 7B Model in Private LLM on iPhone, demonstrating uncensored roleplay interaction — Eva Qwen 2.5 7B Model in Private LLM showcases an immersive roleplay scenario on iPhone, ideal for storytelling and creative dialogues.

Eva Qwen2.5 models cater to various hardware capabilities, offering flexibility for different devices:

Eva D Qwen2.5 1.5B v0.0

Available on devices with 4GB or more RAM, including iPhone 12, iPhone 12 mini, iPhone 13, iPhone 13 mini, and iPhone SE (3rd generation). This model delivers efficient performance with a compact architecture that’s perfect for lighter tasks and on-the-go creativity.

Eva Qwen2.5 7B v0.1

Designed for devices with 8GB or more RAM, including iPhone 15 Pro, iPhone 16, and iPhone 16 Pro. This model balances advanced capabilities and accessibility, making it ideal for immersive roleplaying and storytelling experiences.

Eva Qwen2.5 14B v0.2

For devices with 16GB or more RAM, such as newer-generation iPad Pros with 1TB or more of storage. This model also works seamlessly on Macs with 16GB of RAM, bringing enhanced coherence and depth for long-form creative writing and intricate narratives.

Eva Qwen2.5 32B v0.2

Exclusive to Mac devices with 32GB or more RAM, this model offers top-tier performance for demanding tasks, handling extended contexts with ease and delivering unmatched quality.

System Prompt for Best Roleplaying Results

To achieve optimal performance with EVA Qwen 2.5, use this recommended system prompt:

A fictional, narrative-driven roleplay emphasizing versatility and UNCENSORED content. Adherence to the Role-playing Guidelines is mandatory. Refer to the Role-play Context for accurate information.

Full prompt details are available here.

Ollama and LM Studio vs Private LLM

A significant factor behind Private LLM's superior performance and text generation quality is its use of GPTQ and OmniQuant for quantization. Unlike the traditional Round-to-Nearest (RTN) quantization that Ollama and LM Studio employ, OmniQuant preserves the model's weight distribution more effectively. This results in:

Better Inference Performance: Models quantized with OmniQuant run faster, providing quicker responses without compromising accuracy.
Improved Model Perplexity: OmniQuant maintains higher model fidelity, leading to more coherent and contextually accurate text generation.

In fact, our 3-bit OmniQuant models are competitive with the 4-bit RTN quantized models used by Ollama and others. This means you get similar, if not better, performance and quality in a smaller, more efficient package.

Don’t just take our word for it—compare for yourself.

Explore Eva Qwen 2.5 Today

Experience the latest in local AI with Private LLM v1.9.3 for iPhone and iPad, and v1.9.5 for Mac. Whether you’re crafting a novel, running a tabletop RPG session, or exploring NSFW storytelling, EVA-Qwen2.5 models offer unparalleled creativity, privacy, and performance—all offline.

Download Private LLM now and unlock the potential of uncensored, private AI storytelling on your favorite Apple devices.