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: 10,136 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 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 | from __future__ import annotations
import math
import re
from collections.abc import Collection
from decimal import Decimal, DecimalException
from typing import Any
from jsonschema import Draft202012Validator, ValidationError
from . import ar
COPY_INSTRUCTION = " Copy the complete requested number exactly as printed, including every digit, currency symbol, grouping separator, sign or parentheses. Do not calculate, normalize or round. Return null only if this requested value is absent."
PAYMENT_PRESENCE_INSTRUCTION = " Every separately printed payment actually received is a present row, including an amount of zero, a negative amount, or a few cents. Copy all matching printed received-payment rows in their printed order, preserving duplicate rows. Do not return null or end the list while a matching received-payment row remains."
_NUMBER = re.compile(
"\\s*(?P<open>\\()?\\s*(?P<before>[+-]?)\\s*(?:[$€£¥]|USD|EUR|GBP|CAD|AUD|JPY)?\\s*(?P<after>[+-]?)\\s*(?P<number>(?:\\d{1,3}(?:,\\d{3})+|\\d+)(?:\\.\\d+)?(?:[eE][+-]?\\d+)?)\\s*(?P<close>\\))?\\s*",
re.IGNORECASE,
)
_TRANSFORM = re.compile(
"\\b(?:calculat\\w*|comput\\w*|deriv\\w*|convert\\w*|summar\\w*|aggregat\\w*|group\\w*|combin\\w*|deduplicat\\w*|normaliz\\w*)\\b",
re.IGNORECASE,
)
def _literal(spec: dict[str, Any]) -> bool:
instructions = spec.get("instructions", "")
if not isinstance(instructions, str) or spec.get("knowledge"):
return False
instructions = re.sub(
"\\b(?:do\\s+not|never|don't)\\s+(?:calculate\\w*|compute\\w*|derive\\w*|convert\\w*|summarize\\w*|aggregate\\w*|group\\w*|combine\\w*|deduplicate\\w*|normalize\\w*)\\b",
"",
instructions,
flags=re.IGNORECASE,
)
return not _TRANSFORM.search(instructions)
def _payment_property(spec: dict[str, Any]) -> str | None:
properties = spec.get("properties", {})
if (
spec.get("type") != "object"
or len(properties) != 2
or (not _literal(spec))
or (
not all(
(
isinstance(child, dict) and _literal(child)
for child in properties.values()
)
)
)
or (sum((child.get("type") == "text" for child in properties.values())) != 1)
or (sum((child.get("type") == "number" for child in properties.values())) != 1)
):
return None
return next(
(name for (name, child) in properties.items() if child.get("type") == "number")
)
def payment_lists(fields: dict[str, dict[str, Any]]) -> list[str]:
selected = []
for name, spec in fields.items():
item = spec.get("items")
if (
spec.get("type") != "list"
or not isinstance(item, dict)
or (not _literal(spec))
or (_payment_property(item) is None)
):
continue
positive = re.split(
"[.\\n]|\\bExclude\\b",
spec.get("instructions", ""),
maxsplit=1,
flags=re.IGNORECASE,
)[0]
if (
re.search("\\bpayments?\\b[^.\\n]*\\breceived\\b", positive, re.IGNORECASE)
or re.search(
"\\breceived\\b[^.\\n]*\\bpayments?\\b", positive, re.IGNORECASE
)
) and (
not re.search(
"\\b(?:not|never|exclude|excluding|ignore|without)\\b",
positive,
re.IGNORECASE,
)
):
selected.append(name)
return selected
def normalize(value: object, spec: dict[str, Any]) -> int | float | None:
if value is None:
return None
if not isinstance(value, str):
raise TypeError("A printed number must be a string")
raw = value.strip()
unit = spec.get("unit")
if unit and raw.endswith(unit):
raw = raw[: -len(unit)].rstrip()
match = _NUMBER.fullmatch(raw)
if (
match is None
or bool(match["open"]) != bool(match["close"])
or (match["before"] and match["after"])
or (match["open"] and (match["before"] or match["after"]))
):
raise ValueError("Unsupported printed number")
try:
number = Decimal(match["number"].replace(",", ""))
if not math.isfinite(float(number)):
raise ValueError("Nonfinite numeric answer")
except DecimalException as error:
raise ValueError("Unsupported printed number") from error
if match["open"] or "-" in (match["before"], match["after"]):
number = number.copy_negate()
result = ar.normalize_number(number, spec)
if isinstance(result, float) and (not math.isfinite(result)):
raise ValueError("Nonfinite numeric answer")
Draft202012Validator(ar.schema(spec)).validate(result)
return result
def _text_spec(spec: dict[str, Any]) -> dict[str, Any]:
return {
"type": "text",
"nullable": spec.get("nullable", True),
"instructions": spec.get("instructions", "") + COPY_INSTRUCTION,
}
def prepare(
fields: dict[str, dict[str, Any]], *, payment_rows: Collection[str] = ()
) -> tuple[dict[str, dict[str, Any]], dict[str, dict[str, Any]]]:
fields = {name: dict(spec) for (name, spec) in fields.items()}
contracts = {}
payment_rows = set(payment_rows)
for parent in payment_lists(fields):
prefix = parent + ".item"
children = [
name
for name in payment_rows
if name.startswith(prefix) and name[len(prefix) :].isdigit()
]
if not children:
continue
instructions = fields[parent].get("instructions", "")
if PAYMENT_PRESENCE_INSTRUCTION in instructions:
continue
fields[parent] = {
**fields[parent],
"instructions": instructions + PAYMENT_PRESENCE_INSTRUCTION,
}
for child in children:
if child not in fields or _payment_property(fields[child]) is None:
continue
requested = fields[child].get("instructions", "")
fields[child] = {
**fields[child],
"instructions": requested.replace(
instructions, instructions + PAYMENT_PRESENCE_INSTRUCTION, 1
)
if instructions and instructions in requested
else requested + PAYMENT_PRESENCE_INSTRUCTION,
}
for name, spec in fields.items():
if spec["type"] == "number" and spec.get("source_only"):
contracts[name] = {"scalar": spec}
fields[name] = _text_spec(spec)
elif (
spec["type"] == "list"
and spec.get("source_only")
and (spec["items"] == "number")
):
fields[name] = {
**spec,
"items": "text",
"instructions": spec.get("instructions", "") + COPY_INSTRUCTION,
}
elif (
name in payment_rows
and (number_name := _payment_property(spec)) is not None
):
number_spec = spec["properties"][number_name]
contracts[name] = {"row": spec, "properties": {number_name: number_spec}}
fields[name] = {
**spec,
"properties": {
**spec["properties"],
number_name: _text_spec(number_spec),
},
}
return (fields, contracts)
def _normalize_row(
raw: object, contract: dict[str, Any], record: dict[str, Any], name: str
) -> dict[str, Any] | None:
if raw is None:
result = None
elif isinstance(raw, dict):
result = dict(raw)
current = record.get("fields", {}).get(name, {}).get("properties", {})
for property_name, spec in contract["properties"].items():
value = raw[property_name]
if current.get(property_name, {}).get("type") == "number":
if value is not None and (
isinstance(value, bool)
or not isinstance(value, (int, float, Decimal))
):
raise TypeError("A numeric row value must be a number")
result[property_name] = ar.normalize_number(value, spec)
else:
if value is not None and (not record.get("_media_payload")):
if not isinstance(value, str) or value not in record["input"]:
raise ValueError("not_in_source")
result[property_name] = normalize(value, spec)
else:
raise TypeError("A printed payment row must be an object")
Draft202012Validator(ar.schema(contract["row"])).validate(result)
return result
def normalize_predictions(
record: dict[str, Any], predictions: dict[str, dict[str, Any]]
) -> dict[str, dict[str, Any]]:
result = dict(predictions)
for name, contract in record.get("_meta", {}).get("printed_numbers", {}).items():
answer = dict(result[name])
result[name] = answer
if not answer.get("valid"):
continue
raw = answer["value"]
try:
if "properties" in contract:
answer["value"] = _normalize_row(raw, contract, record, name)
else:
if raw is not None and (not record.get("_media_payload")):
if not isinstance(raw, str) or raw not in record["input"]:
raise ValueError("not_in_source")
answer["value"] = normalize(raw, contract["scalar"])
answer["printed_value"] = raw
answer["normalization_method"] = "printed_number_v1"
except (
ValueError,
TypeError,
KeyError,
OverflowError,
ValidationError,
) as error:
answer.update(
value=None,
valid=False,
confidence=0.0,
span=None,
failure_reason="not_in_source"
if str(error) == "not_in_source"
else "invalid_printed_number",
)
return result
|