Pairs an iPhone with a Mac over USB-C so the phone's GPU helps run a Qwen 27B model locally, speeding up prompt reading and extending context.
hr98w/jev-visual
A small educational project for running vision-language model inference on Apple Silicon with MLX, scoring candidate answers about one image.
- スター
- 319
- フォーク
- 0
- 未解決の Issue
- 0
- 最終コミット
- 29 日前
主な機能
- 01Asks several questions about one image in a single shared-context pass.
- 02Scores candidate answers directly from model logits instead of generating JSON.
- 03Provides a local browser UI, CLI and HTTP API.
- 04Includes visual demos such as a sorting game, Breakout variant and camera gesture recognition.
- 05Ships benchmarks comparing generation-based and shared-prefix scoring.
ユースケース
- Learning how vision-language models can be used for classification without structured generation.
- Experimenting with multi-question image analysis on a local Apple Silicon Mac.
- Studying KV cache reuse and candidate scoring in MLX.
- Prototyping camera-based interactions with a small local model.
トピック
よくある質問
- jev-visual とは?
- A small educational project for running vision-language model inference on Apple Silicon with MLX, scoring candidate answers about one image.
- jev-visual は無料で使えますか?ライセンスは?
- jev-visual は MIT ライセンスのオープンソースです。商用製品で使う前にライセンス条件を確認してください。
- jev-visual は今もメンテナンスされていますか?
- 活発にメンテナンス中。jev-visual の最終コミットは 2026年9月21日 です。
- jev-visual は何に使えますか?
- Learning how vision-language models can be used for classification without structured generation. Experimenting with multi-question image analysis on a local Apple Silicon Mac. Studying KV cache reuse and candidate scoring in MLX. Prototyping camera-based interactions with a small local model.
- jev-visual はどこで入手できますか?
- ソースコードとインストール手順は GitHub にあります:https://github.com/hr98w/jev-visual
概要はプロジェクトの README をもとに AI が生成しています。GitHub の数値は毎日更新されます。
関連プロジェクト
Swift and Metal runtime that runs the 26B Gemma 4 26B-A4B model on Apple Silicon Macs with about 2 GB of RAM by streaming experts from SSD.
Open-source native unified multimodal models from SenseNova that handle image understanding, text-to-image generation, and image editing in one architecture.
Runs open AI models locally on Apple Silicon Macs via MLX and serves them through an Anthropic-API-compatible server so Claude Code works offline.
A 9B-parameter open-weight multimodal model for visual understanding, spatial reasoning, agentic tool use, and embodied-AI research, built on Qwen3.5 and C-RADIOv4-H.
Open-source native vision-language models (NEO series) that process pixels and words in one encoder-free dense architecture, with 2B and 9B releases.