moss-image-base / README.md
infosave's picture
MOSS-Image-Base CMF: CPU Space via cortiq v0.8.31
ecf4609 verified
|
Raw History Blame Contribute Delete
1.16 kB
metadata
title: MOSS-Image-Base on cortiq
emoji: 🖼️
colorFrom: indigo
colorTo: purple
sdk: docker
app_port: 7860
pinned: false
license: apache-2.0
short_description: MOSS-Image-Base via cortiq — CPU text-to-image
tags:
  - cmf
  - moss-image
  - text-to-image

MOSS-Image-Base on cortiq

This Space runs cortiq against infosave/MOSS-Image-Base-cmf — the MOSS Base text-to-image path (Qwen3.5 encoder + 42-layer MOSSImageTransformer + FLUX.2 VAE) in one memory-mapped CMF file.

app.py collects a prompt and shells out to cortiq imagine. There is no PyTorch, no diffusers and no model code in this Space at all. CPU-only: the Space is honest about latency (see the model card Vulkan baseline for GPU numbers).

  • moss-image-base-q8_2f.cmf — 12.35 GiB, quality-first (q8_2f DiT)
  • moss-image-base-q4tp.cmf — 10.11 GiB, lower DiT footprint (q4tp; evaluate on your prompts)

Quality recipe: 1024×1024, 50 steps, CFG 3, shift 5, empty negative prompt, negative-branch skip at layer 22. On CPU try 512×512 / 8 steps for a smoke first.