A small model that answers questions about a person from a fact sheet in the system prompt. It runs in the browser with WebLLM and WebGPU.
The model does not store facts about any person. Give it a different fact sheet and it answers about that person. It is a general model. It is not the model on yipengsun.com.
import * as webllm from "https://cdn.jsdelivr.net/npm/@mlc-ai/web-llm@0.2.86/lib/index.js";
const repo = "https://huggingface.co/yipengsun/grounded-persona-Qwen3.5-0.8B-q4f16_1-MLC/resolve/main/";
const engine = await webllm.CreateMLCEngine("grounded-persona", {
appConfig: { model_list: [{
model: repo,
model_id: "grounded-persona",
model_lib: repo + "qwen3_5-0.8b-q4f16_1-vocab51200-webgpu.wasm",
overrides: { context_window_size: 4096 },
}] },
});
const reply = await engine.chat.completions.create({
messages: [{ role: "system", content: RULES + "\n" + FACT_SHEET }, ...history],
temperature: 0,
extra_body: { enable_thinking: false },
});
SITE FACTS:. Keep it under about 2,000 tokens.RULES tells the model to answer in the first person, use only the sheet, and refuse anything else with the sheet's public email.A 0.8B model can make mistakes. It can answer only what the fact sheet contains.
Apache 2.0, the same as the base model.
params_shard_4.bin 33 MBparams_shard_5.bin 33 MBparams_shard_7.bin 33 MBparams_shard_9.bin 33 MBparams_shard_8.bin 33 MBparams_shard_2.bin 32 MBparams_shard_3.bin 32 MBparams_shard_6.bin 31 MBparams_shard_0.bin 26 MBparams_shard_1.bin 21 MBparams_shard_10.bin 4 MBQwen/Qwen3.5-0.8B