gw0/jobfit-model

A CV/job-post fit-scoring typed-decision model, training pipeline, and web app -- JobFit. Try it in the HF Space with model weights from HF Model.

This model is a Jev-shaped typed-decision model: a state (CV and job description) plus typed questions (score / choice / noul) in, one typed answer per question id out, each read off a causal LM's own lm_head in one masked forward pass. It answers the 17 JobFit questions with a 5-level score and a calibrated confidence.

• base model: HuggingFaceTB/SmolLM2-135M-Instruct
• git SHA: bf25c05e6deb457c5a1e3f0842d6dc912325aec8
• mean MAE (quantized, test): 0.1568
• beats train-mean floor on 15/17 questions

Usage

Ships as dynamic-int8 ONNX for transformers.js, loaded through the generic PreTrainedModel class. The graph takes input_ids, segment_ids, answer_positions and candidate_ids and returns answer_logits; calibration.json holds the model's temperature and JobFit's insufficient-data confidence threshold.

import { AutoTokenizer, PreTrainedModel } from "@huggingface/transformers";
const tokenizer = await AutoTokenizer.from_pretrained("gw0/jobfit-model");
const model = await PreTrainedModel.from_pretrained("gw0/jobfit-model", { dtype: "q8" });

The input encoding and answer decoding are in the JobFit source (frontend/src/jev.mjs, frontend/src/infer.ts).

Limitations

A research benchmark, not a hiring tool: don't use it as the sole basis for decisions about people. Training CVs are synthetic and labels come from an LLM judge. The full benchmark report is in the repository.

License

Copyright © 2026 gw0 <gw.2026@ena.one>. Licensed under AGPL-3.0-or-later.

The weights on the HF Model carry their base model's licence (see the model card).

CVs are synthetic. Job posts come from public Greenhouse, Lever, Ashby and Workday pages, keep their source url, and remain under their publishers' rights and licenses. Not a legal opinion.

If my work saves you time, money, or tokens, consider sponsoring the work that keeps it maintained — and tell whoever else benefits, human or agent.