---
title: Occamy Checkpoint Explorer
emoji: 🐉
colorFrom: green
colorTo: blue
sdk: static
app_file: index.html
pinned: true
license: apache-2.0
models:
- Accio-Lab/occamy-1.0
- Accio-Lab/occamy-1.0-GGUF
- Accio-Lab/occamy-1.0-APEX-GGUF
- Accio-Lab/occamy-1.0-FP8
- Accio-Lab/occamy-1.0-NVFP4
- Accio-Lab/occamy-1.0-MLX-8bit
- Accio-Lab/occamy-1.0-MLX-6bit
- Accio-Lab/occamy-1.0-MLX-5bit
- Accio-Lab/occamy-1.0-MLX-mxfp8
- Accio-Lab/occamy-1.0-MLX-mxfp4
- Accio-Lab/occamy-1.0-MLX-nvfp4
- Accio-Lab/occamy-1.0-MLX-4bit
- Accio-Lab/occamy-1.0-MLX-3bit
- Accio-Lab/occamy-1.0-MTP
short_description: Compare Occamy weights and copy deployment commands.
---

# Occamy Checkpoint Explorer

The official Occamy model selection, download and deployment guide. Compare twenty-four GGUF options (seventeen standard recipes and seven APEX profiles, including calibrated variants and Mini), BF16/FP8/NVFP4 server checkpoints, MLX 3/4/5/6/8-bit and MXFP4/MXFP8/NVFP4 candidates and the experimental MTP head.

This static Space runs in the browser with no inference backend or paid compute. It provides links and deployment recipes; it does not run the 35B model or download weights automatically.

Sizes come from inspected Hub file metadata. The optional weight-budget filter checks file sizes only: KV cache, runtime buffers, images and operating-system memory require additional space. Runtime claims and validation limits are shown for each selection. Mac Metal acceptance of the MLX candidates is pending.

See the [Occamy collection](https://huggingface.co/collections/Accio-Lab/occamy-10-6ac00729659e8a86ec11e216), [MLX collection](https://huggingface.co/collections/Accio-Lab/occamy-10-mlx-6ac0072a7c1cdbc5e418e90b), [model card](https://huggingface.co/Accio-Lab/occamy-1.0), and [paper](https://arxiv.org/abs/2609.11977).

MLX selections offer a single prompt recipe or a local API server, with a separate curl request and client base URL. Native Linux checks for the new 5-bit and floating-format exports include eight authored inference fixtures, two HTTP API checks and paired held-out BF16 subset PPL. Full APEX calibration coverage and measured per-profile quality are documented in its model card. Mac Metal acceptance remains pending.

## Related paper

This deployment guide accompanies [Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work](https://arxiv.org/abs/2609.11977). The [released training data](https://huggingface.co/datasets/Accio-Lab/occamy-data-1.0) has the full paper citation.

The visual style follows the [current project website](https://accio-lab.github.io/occamy/): white background, Inter text, a Newsreader wordmark and the same green accent.