Dolphin 3.0 is an uncensored, steerable instruction-tuned model based on the Qwen 2.5 architecture. This version allows for private, high-performance inference directly in the browser or Node.js using WebGPU or WASM.

Model Details

Model Type: Causal Language Model
Architecture: Qwen2.5
Parameters: 1.54 Billion
Format: ONNX
Base Model: dphn/Dolphin3.0-Qwen2.5-1.5B

Usage with Transformers.js

Basic Example (WebGPU)

javascript

import { pipeline } from '@xenova/transformers';

const generator = await pipeline('text-generation', 'd1g1mortal/Dolphin3.0-Qwen2.5-1.5B-ONNX', {
    device: 'webgpu', 
});

const messages = [
    { role: "system", content: "You are Dolphin, a helpful AI assistant." },
    { role: "user", content: "Write a short creative poem about the ocean." },
];

const prompt = generator.tokenizer.apply_chat_template(messages, {
    tokenize: false,
    add_generation_prompt: true,
});

const output = await generator(prompt, {
    max_new_tokens: 256,
    temperature: 0.7,
    do_sample: true,
});

console.log(output[0].generated_text);