You can deploy this model and start earning money today!

LJTSG/
xLAM-2-3b-fc-r-q4f16_1-MLC

Name: xLAM-2-3b-fc-r-q4f16_1-MLC
Availability: InStock

Salesforce/xLAM-2-3b-fc-r

quantized

cc-by-nc-4.0

Views

No views yet

xLAM-2-3b-fc-r — q4f16_1 MLC (WebLLM / WebGPU)

An MLC-LLM compile of Salesforce/xLAM-2-3b-fc-r — Salesforce's function-calling ("fc-r") small model — quantized to q4f16_1 so it runs entirely in the browser on WebGPU via WebLLM. No server, no API.

To our knowledge this is the first xLAM-2 build published in WebLLM/MLC format.

xLAM-2-3b is a Qwen2.5-3B architecture model, so it reuses the prebuilt Qwen2.5-3B WebGPU model library — no separate .wasm compile needed.

Use in WebLLM

import { CreateMLCEngine } from "@mlc-ai/web-llm";

const appConfig = {
  model_list: [{
    model: "https://huggingface.co/LJTSG/xLAM-2-3b-fc-r-q4f16_1-MLC",
    model_id: "xLAM-2-3b-fc-r-q4f16_1-MLC",
    model_lib:
      "https://raw.githubusercontent.com/mlc-ai/binary-mlc-llm-libs/main/web-llm-models/" +
      "v0_2_84/base/Qwen2.5-3B-Instruct-q4f16_1_cs1k-webgpu.wasm",
  }],
};

const engine = await CreateMLCEngine("xLAM-2-3b-fc-r-q4f16_1-MLC", { appConfig });

// xLAM is trained for tool-calling — pass tools and it emits structured calls:
const res = await engine.chat.completions.create({
  messages: [{ role: "user", content: "What's the weather in Paris?" }],
  tools: [{ type: "function", function: {
    name: "get_weather",
    parameters: { type: "object", properties: { city: { type: "string" } }, required: ["city"] },
  }}],
  tool_choice: "auto",
});

Details

Quantization: q4f16_1 (group-quant, group_size 32, int4) — ~1.8 GB, ~2.5 GB VRAM in WebGPU
Conversion: mlc_llm convert_weight + gen_config --conv-template qwen2, built from source (matched TVM-unity + mlc-llm)
62 param shards + tokenizer (byte-level BPE)

License

Inherits the license of the base model, Salesforce/xLAM-2-3b-fc-r. Non-commercial (CC-BY-NC-4.0). Use accordingly.