The model was exported with Liquid AI's official LiquidONNX tooling at revision 9a23ddd23035165f7414a5de3220a51e85780f64. Q4 uses symmetric quantization for WebGPU compatibility.

The portfolio supplies focused verified profile context and recent conversation history at inference time. This checkpoint is fine-tuned for privacy boundaries, visitor-identity handling, career chronology, and multi-turn product questions; those answers are generated by the model rather than returned as fixed strings.

javascript

import { pipeline } from "@huggingface/transformers";

const generator = await pipeline(
  "text-generation",
  "danelcsb/daniel-lfm2-350m-ONNX",
  { device: "webgpu", dtype: "q4" },
);