clairic-flash / README.md
multimodalart's picture
multimodalart HF Staff
Upload folder using huggingface_hub
e938d78 verified
|
Raw History Blame Contribute Delete
992 Bytes
---
title: Clairic Flash
emoji: ⚡
colorFrom: yellow
colorTo: blue
sdk: gradio
sdk_version: 6.32.0
app_file: app.py
short_description: Cloudflare Clairic Flash 9B typed structured extraction
python_version: "3.12"
startup_duration_timeout: 1h
models:
- Cloudflare/clairic-flash
---
# Clairic Flash · System One: Extract
Gradio demo for [Cloudflare/clairic-flash](https://huggingface.co/Cloudflare/clairic-flash), a 9B multimodal
model (Qwen3.5-9B base + typed output head) that turns a text/JSON state (and optional images) plus a schema of
typed fields into structured, scored answers (confidence, p_null, optional quantiles).
Runs on ZeroGPU. The release runtime's Hopper-only kernels (FA3, DeepGEMM, FlashInfer RMSNorm, CUDA-graph
decode) are replaced by SDPA and numerically equivalent PyTorch ops so it runs on Blackwell sm_120.
Example receipt images come from the [CORD-v2 dataset](https://huggingface.co/datasets/naver-clova-ix/cord-v2)
by NAVER Clova AI, licensed CC-BY-4.0.