Image-Text-to-Text
Transformers
Safetensors
qwen3_5
clairic
cloudflare
systemone
multimodal
structured-output
information-extraction
custom-code
conversational
Instructions to use Cloudflare/clairic-flash with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Cloudflare/clairic-flash with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="Cloudflare/clairic-flash") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("Cloudflare/clairic-flash") model = AutoModelForMultimodalLM.from_pretrained("Cloudflare/clairic-flash", device_map="auto") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use Cloudflare/clairic-flash with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Cloudflare/clairic-flash" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Cloudflare/clairic-flash", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker
docker model run hf.co/Cloudflare/clairic-flash
- SGLang
How to use Cloudflare/clairic-flash with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "Cloudflare/clairic-flash" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Cloudflare/clairic-flash", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "Cloudflare/clairic-flash" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Cloudflare/clairic-flash", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }' - Docker Model Runner
How to use Cloudflare/clairic-flash with Docker Model Runner:
docker model run hf.co/Cloudflare/clairic-flash
File size: 3,178 Bytes
dc7d8aa | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 | """Conservative page orientation from detached dots beside Latin letter stems.
This examines glyph geometry only; it performs no text readout. Pages without
enough consistent evidence retain their supplied orientation.
"""
from __future__ import annotations
from typing import TYPE_CHECKING
if TYPE_CHECKING:
from PIL import Image
def document_rotation(image: Image.Image) -> tuple[int, list[int]]:
import cv2
import numpy as np
page = image.convert("L")
page.thumbnail((1600, 1600))
pixels = np.asarray(page)
_, ink = cv2.threshold(pixels, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)
_, _, stats, _ = cv2.connectedComponentsWithStats(ink, connectivity=8)
stats = stats[1:]
votes = [0, 0, 0, 0]
if not len(stats) or len(stats) > 12_000:
return 0, votes
width, height, area = stats[:, 2:].T
short, long = np.minimum(width, height), np.maximum(width, height)
stems = stats[(long >= 9) & (long <= 60) & (long >= 2.5 * short) & (area >= 8)]
dots = stats[(width <= 12) & (height <= 12) & (area >= 2) & (area <= 100)]
if not len(stems) or not len(dots) or len(stems) > 1200 or len(dots) > 2400:
return 0, votes
dx, dy, dw, dh, da = dots.T
for sx, sy, sw, sh, sa in stems:
vertical = sh > sw
across = dx + dw / 2 if vertical else dy + dh / 2
center = sx + sw / 2 if vertical else sy + sh / 2
start, length = (sy, sh) if vertical else (sx, sw)
left, size = (dy, dh) if vertical else (dx, dw)
thin = sw if vertical else sh
aligned = (abs(across - center) <= max(1.5, thin * 0.6)) & (
size <= length * 0.4
)
aligned &= (dw <= max(3, thin * 2)) if vertical else (dh <= max(3, thin * 2))
aligned &= da < sa * 0.7
before = (
aligned & (left + size <= start) & (start - left - size <= length * 0.45)
)
after = (
aligned
& (left >= start + length)
& (left - start - length <= length * 0.45)
)
if vertical:
votes[0] += int(np.any(before))
votes[2] += int(np.any(after))
else:
votes[1] += int(np.any(before))
votes[3] += int(np.any(after))
letters = (area >= 10) & (short >= 3) & (long <= 60) & (long <= 4 * short)
if int(letters.sum()) < 80:
return 0, votes
aspect = float(np.median(width[letters] / height[letters]))
if aspect < 0.8:
directions = (0, 2)
elif aspect > 1.25:
directions = (1, 3)
else:
return 0, votes
# Text-line direction excludes accidental dot/stem pairs in the other axis.
winner = max(directions, key=votes.__getitem__)
runner = votes[directions[0] if winner == directions[1] else directions[1]]
if (
votes[winner] < 12
or votes[winner] < 3 * max(1, runner)
or votes[winner] < 0.7 * (votes[winner] + runner)
):
return 0, votes
return winner * 90, votes
def orient_document(image: Image.Image) -> Image.Image:
rotation, _ = document_rotation(image)
return image.rotate(-rotation, expand=True) if rotation else image
|