Image-Text-to-Text
Transformers
Safetensors
qwen3_5
clairic
cloudflare
systemone
multimodal
structured-output
information-extraction
custom-code
conversational
Instructions to use Cloudflare/clairic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Cloudflare/clairic with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="Cloudflare/clairic") 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") model = AutoModelForMultimodalLM.from_pretrained("Cloudflare/clairic", 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 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Cloudflare/clairic" # 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", "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
- SGLang
How to use Cloudflare/clairic 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" \ --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", "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" \ --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", "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 with Docker Model Runner:
docker model run hf.co/Cloudflare/clairic
File size: 6,448 Bytes
f65336d | 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 | from __future__ import annotations
import re
from datetime import datetime
from typing import Any
TYPES = ("date", "time", "datetime")
COPY_INSTRUCTION = " Copy only a date or time explicitly printed for this exact requested role. Return null if this role has no printed date or time. Do not use invoice or due dates for service or billing period boundaries. Do not infer a period from its duration. Copy the requested date or time exactly as printed, including AM or PM when present. Preserve every digit. Do not normalize or convert the value. Return null only if the requested value is absent."
_LITERAL = re.compile("\\b(?:print\\w*|explicit\\w*|verbatim|shown)\\b", re.IGNORECASE)
_DERIVED = re.compile(
"\\b(?:calculat\\w*|comput\\w*|deriv\\w*|infer\\w*|after|before|ago|next|previous|following|prior|add|subtract|plus|minus|earlier|later|offset|duration|terms)\\b|\\bnet\\s*\\d+\\b|\\b(?:first|last)\\s+day\\s+of\\b",
re.IGNORECASE,
)
_DOCUMENT_DATE = re.compile(
"\\b(?:invoice|billing|service|subscription|document|receipt)\\b", re.IGNORECASE
)
_DATE_ROLES = {
"invoice_date",
"due_date",
"period_start",
"period_end",
"start_date",
"end_date",
}
def explicit_fields(fields: dict[str, dict[str, Any]]) -> dict[str, dict[str, Any]]:
result = dict(fields)
for name, spec in fields.items():
date_list = spec.get("type") == "list" and spec.get("items") == "date"
if (
spec.get("type") not in TYPES
and (not date_list)
or "source_only" in spec
or "null_threshold" in spec
or spec.get("knowledge")
):
continue
instructions = spec.get("instructions", "")
if not isinstance(instructions, str):
continue
role = name.casefold() in _DATE_ROLES and _DOCUMENT_DATE.search(instructions)
if (_LITERAL.search(instructions) or role) and (
not _DERIVED.search(instructions)
):
result[name] = {**spec, "source_only": True}
return result
def clock(value: str) -> str:
match = re.fullmatch(
"\\s*(\\d{1,2}):(\\d{2})(?::(\\d{2}))?\\s*([AP]\\.?M\\.?)?\\s*",
value,
re.IGNORECASE,
)
if match is None:
raise ValueError("Unsupported printed time")
(hour, minute, second) = map(int, (match[1], match[2], match[3] or "0"))
meridiem = match[4]
if meridiem:
if not 1 <= hour <= 12:
raise ValueError("Invalid 12-hour clock")
hour = hour % 12 + (12 if meridiem.replace(".", "").upper() == "PM" else 0)
if not 0 <= hour < 24 or not 0 <= minute < 60 or (not 0 <= second < 60):
raise ValueError("Invalid printed time")
return f"{hour:02}:{minute:02}:{second:02}"
def day(value: str) -> str:
value = value.strip()
for fmt in (
"%Y-%m-%d",
"%Y/%m/%d",
"%Y.%m.%d",
"%B %d, %Y",
"%b %d, %Y",
"%B %d %Y",
"%b %d %Y",
"%d %B %Y",
"%d %b %Y",
):
try:
return datetime.strptime(value, fmt).date().isoformat()
except ValueError:
pass
numeric = re.fullmatch("(\\d{1,2})([-/.])(\\d{1,2})\\2(\\d{4})", value)
if numeric:
(first, second, year) = (int(numeric[1]), int(numeric[3]), int(numeric[4]))
readings = set()
for month, date in ((first, second), (second, first)):
try:
readings.add(datetime(year, month, date).date().isoformat())
except ValueError:
pass
if len(readings) == 1:
return readings.pop()
raise ValueError("Unsupported or ambiguous printed date")
def normalize(value: object, kind: str) -> str | None:
if value is None:
return None
if not isinstance(value, str):
raise TypeError("A printed temporal answer must be a string")
if kind == "time":
return clock(value)
if kind == "date":
return day(value)
if kind == "datetime":
match = re.fullmatch(
"(.+?)(?:T|\\s+(?:at\\s+)?)(\\d{1,2}:\\d{2}(?::\\d{2})?\\s*(?:[AP]\\.?M\\.?)?)",
value.strip(),
re.IGNORECASE,
)
if match is None:
raise ValueError("Unsupported printed datetime")
return day(match[1]) + "T" + clock(match[2])
raise ValueError("Unsupported temporal type")
def prepare(
fields: dict[str, dict[str, Any]],
) -> tuple[dict[str, dict[str, Any]], dict[str, str]]:
fields = {name: dict(spec) for (name, spec) in fields.items()}
kinds = {}
for name, spec in fields.items():
if not spec.get("source_only"):
continue
if spec["type"] == "list" and spec["items"] == "date":
spec["items"] = "text"
spec["instructions"] = spec.get("instructions", "") + COPY_INSTRUCTION
continue
kind = spec["type"]
if kind not in TYPES:
continue
kinds[name] = kind
fields[name] = {
"type": "text",
"nullable": spec.get("nullable", True),
"instructions": spec.get("instructions", "") + COPY_INSTRUCTION,
}
return (fields, kinds)
def normalize_predictions(
record: dict[str, Any], predictions: dict[str, dict[str, Any]]
) -> dict[str, dict[str, Any]]:
result = dict(predictions)
for name, kind in record.get("_meta", {}).get("printed_temporal", {}).items():
answer = dict(result[name])
result[name] = answer
if not answer.get("valid"):
continue
printed = answer["value"]
answer["printed_value"] = printed
if (
printed is not None
and (not record.get("_media_payload"))
and (not isinstance(printed, str) or printed not in record["input"])
):
answer.update(
value=None,
valid=False,
confidence=0.0,
span=None,
failure_reason="not_in_source",
)
continue
try:
answer["value"] = normalize(printed, kind)
answer["normalization_method"] = "printed_temporal_v1"
except (TypeError, ValueError):
answer.update(
value=None,
valid=False,
confidence=0.0,
failure_reason="invalid_printed_temporal",
)
return result
|