google/gemma-4-E2B-it in q4f16_1 for browser-local WebGPU and MLC-LLM runtimes.mlc-llm / TVM fork and reflects the cleaned baseline validated on 2026-04-13.prefill_chunk_size=16 workaround is no longer needed.q4f16_1webgpugemma4gemma_instruction1024libs/gemma-4-E2B-it-q4f16_1-MLC-webgpu.wasm: validated WebGPU model librarymlc-chat-config.json: MLC runtime configurationparams_shard_*.bin: quantized parameter shardstensor-cache.json: tensor metadata cachetokenizer.json, tokenizer_config.json: tokenizer assetsrelease-manifest.json: file inventory with SHA-256 hashesmlc_llm chat HF://welcoma/gemma-4-E2B-it-q4f16_1-MLC1import { CreateMLCEngine } from "@mlc-ai/web-llm";
2
3const repo = "https://huggingface.co/welcoma/gemma-4-E2B-it-q4f16_1-MLC";
4
5const appConfig = {
6 model_list: [
7 {
8 model: repo,
9 model_id: "gemma-4-E2B-it-q4f16_1-MLC",
10 model_lib: `${repo}/resolve/main/libs/gemma-4-E2B-it-q4f16_1-MLC-webgpu.wasm`,
11 required_features: ["shader-f16"],
12 },
13 ],
14};
15
16const engine = await CreateMLCEngine("gemma-4-E2B-it-q4f16_1-MLC", {
17 appConfig,
18});mlc-ai release. The validated scope is Gemma 4 E2B text generation on WebGPU.