| --- |
| license: cc-by-nc-4.0 |
| base_model: openjev/OpenJev-Flash-9B |
| tags: |
| - decision-model |
| - zero-shot-classification |
| - mlx |
| - apple-silicon |
| --- |
| |
| # OpenJev Flash 9B, MLX 4-bit (Apple silicon, text only) |
|
|
| This is the language model of [OpenJev Flash 9B](https://huggingface.co/openjev/OpenJev-Flash-9B), converted for Apple silicon with mlx-lm (4-bit, group size 64). It is about 5.0 GB, half the size of the [8-bit build](https://huggingface.co/openjev/OpenJev-Flash-9B-MLX). The main card covers the model, the API and all results. This page says what this build is, how it was checked and how to run it. |
|
|
| **Text only.** The converter keeps the language model and drops the vision tower, so this build answers questions about text, JSON and DOM, not screenshots. |
|
|
| ## Results |
|
|
| On JevBench's 231 public items this 4-bit build scores 186, two items below the 8-bit build: |
|
|
| | model | correct | accuracy | |
| |---|---|---| |
| | Clef-Flash (Cloudflare) | 190 of 231 | 82.3% | |
| | OpenJev Flash 9B, MLX 8-bit | 188 of 231 | 81.4% | |
| | **OpenJev Flash 9B, MLX 4-bit (this build)** | **186 of 231** | **80.5%** | |
| | Kev-9B | 183 of 231 | 79.2% | |
| | Nimble 9B (Bespoke Labs) | 183 of 231 | 79.2% | |
|
|
| No JevBench item was used to train or tune it. More results are on the [main card](https://huggingface.co/openjev/OpenJev-Flash-9B). |
|
|
| ## Run it |
|
|
| ```bash |
| uv venv mlx --python 3.12 |
| uv pip install --python mlx/bin/python "mlx==0.32.2" "mlx-lm==0.31.3" \ |
| "transformers==5.17.0" "openai==3.16.2" "httpx==0.28.1" "huggingface-hub==1.32.0" |
| |
| # Download the text-only weights and the two helper files. |
| mlx/bin/hf download openjev/OpenJev-Flash-9B-MLX-4bit --local-dir OpenJev-Flash-9B-MLX-4bit |
| mlx/bin/hf download openjev/OpenJev-Flash-9B helper/shim.py helper/shim_mlx.py \ |
| --local-dir openjev-flash-9b-api |
| |
| TOKENIZER=OpenJev-Flash-9B-MLX-4bit SHIM_MODEL=OpenJev-Flash-9B-MLX-4bit \ |
| READOUT_T=1.07 READOUT_NOUL_T=1.074766 READOUT_NOUL_BIAS=0 \ |
| READOUT_TARGETED=1 READOUT_INSTR_STYLE=pyrepr SHIM_STAGGER=1 \ |
| mlx/bin/python openjev-flash-9b-api/helper/shim_mlx.py \ |
| --helper openjev-flash-9b-api/helper/shim.py --model OpenJev-Flash-9B-MLX-4bit --port 3000 |
| ``` |
|
|
| Then call `http://localhost:3000/v1/systemone` exactly as the main card describes. `shim_mlx.py` replaces only the helper's model client with an MLX one; prompts, option layout, readout and calibration are the helper's own code. The helper's built-in defaults belong to OpenJev 27B, so pass the five `READOUT_*` settings above. Leave `--prefix-cache` off to reproduce the measured path. Append `--selfcheck` for a quick smoke test that prints the helper hash (it should begin `81a22f1b`), the calibration and a few answers, then exits. |
|
|
| ## Licence |
|
|
| Weights: CC BY-NC 4.0 (research and non-commercial use), the same as the main repository. For a commercial licence, email [support@loopai.com](mailto:support@loopai.com). Helper and serving files: Apache 2.0. The base model, Qwen/Qwen3.5-9B, is Apache 2.0. The licence texts and the base model's attribution are in the main repository's `LICENSE`, `LICENSE-APACHE-2.0` and `NOTICE`. |
|
|
| OpenJev is an independent project, not affiliated with TypeSafe; Jev is their product. |
|
|
