Modelpublic

nicolasembleton/openjev-minicpm5-2b-lora-phase1-onnx

sourceHugging Faceapache-2.0updated 3d agoView on Hugging Face

0likes171downloads

Model Card

Open-Jev Phase-1 MiniCPM5-2B LoRA (Transformers.js ONNX)

Merged Phase-1 LoRA into openbmb/MiniCPM5-2B, exported for browser use with `@huggingface/transformers` AutoModelForCausalLM.

Parents

—Base: `openbmb/MiniCPM5-2B`
—LoRA adapter: `nicolasembleton/openjev-minicpm5-2b-lora-phase1`

Phase-1 eval (PyTorch LoRA; browser ONNX may differ slightly)

—Test overall: 0.467 → 0.702
—OOD overall: 0.473 → 0.710

Layout

Transformers.js convention (like onnx-community/Qwen2.5-0.5B-Instruct):

config.json, tokenizer files
onnx/model.onnx            # fp32 baseline (may include external data)
onnx/model_q4.onnx         # browser default
onnx/model_q4f16.onnx      # optional
onnx/model_fp16.onnx       # optional
quantize_config.json

Load in Transformers.js

import { AutoModelForCausalLM, AutoTokenizer } from "@huggingface/transformers";

const model_id = "nicolasembleton/openjev-minicpm5-2b-lora-phase1-onnx";
const tokenizer = await AutoTokenizer.from_pretrained(model_id);
const model = await AutoModelForCausalLM.from_pretrained(model_id, {
  dtype: "q4",
  device: "webgpu", // or "wasm"
});

Not Nick's older custom layout at nicolasembleton/MiniCPM5-2B-onnx (fp16 prefill+decode).