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: 8,114 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 | from __future__ import annotations
from typing import TYPE_CHECKING
from . import media_state_prompt, po_image_supplement, text_ar, text_list_completion
from . import grounding, identifier_spelling
from .backend import ClairicFlashModel
from .media import media_warmup_groups
if TYPE_CHECKING:
from typing import Any
from .api import PredictRequest
from .backend import EncodedRequest
class Clairic27BModel(ClairicFlashModel):
RELEASE_MODEL_ID = "clairic-27b"
COMPACT_LIST_PROMPTS = False
DATE_ROLE_GUIDANCE = False
SOURCE_ROW_ELIGIBILITY = True
FINANCIAL_SERVICE_GUIDANCE = False
FINANCIAL_FEE_GUIDANCE = True
SPELL_VISUAL_PO_REFERENCES = True
FOCUS_SPARSE_IMAGES = False
PRESERVE_IMAGE_ENTITY_NAMES = True
FOCUS_IMAGE_PARTY_NAMES = False
LITERAL_IMAGE_PAYMENT_ROWS = True
BIND_IMAGE_FINANCIAL_LINES = False
IMAGE_TABLE_ROW_SCALE = 4
BIND_IMAGE_RAGGED_ROWS = False
def encode(self, request: PredictRequest, **options: Any) -> EncodedRequest:
supplement_disabled = options.pop("po_supplement_disabled", False)
encoded = self._encode_without_po_supplement(request, **options)
detail = any(
(
options.get(name)
for name in (
"identifier_view",
"party_name_view",
"physical_row_view",
"identifier_spelling_disabled",
"pdf_spelling_view",
"po_detail_view",
)
)
)
if (
type(self) is not Clairic27BModel
or supplement_disabled
or detail
or (not po_image_supplement.selected_fields(request, encoded))
):
return encoded
if (
self.settings.model_id != "clairic-27b"
or self.max_concurrency != self.settings.max_concurrency
or self.max_concurrency != 1
):
return encoded
return po_image_supplement.prepare(
request,
encoded,
encode_child=lambda child: self.encode(child, po_supplement_disabled=True),
tokenizer=self.tokenizer,
max_length=self.settings.max_length,
media_groups=media_warmup_groups,
)
def _encode_without_po_supplement(
self, request: PredictRequest, **options: Any
) -> EncodedRequest:
detail = any(
(
options.get(name)
for name in (
"identifier_view",
"party_name_view",
"physical_row_view",
"identifier_spelling_disabled",
"pdf_spelling_view",
"po_detail_view",
)
)
)
if type(self) is Clairic27BModel and options.get("po_detail_view"):
options = {**options, "pdf_spelling_view": True}
encoded = super().encode(request, **options)
if type(self) is not Clairic27BModel or detail:
return encoded
(record, forecasts, tasks, typed, eos) = encoded
active = {name for (name, _, _) in tasks}
contracts = record.get("_meta", {}).get("identifier_spelling", {})
po_fields = {
name: request.fields[name]
for (name, contract) in contracts.items()
if name in active
and contract.get("wire") == identifier_spelling.VERSION
and (len(request.images) in (1, 2))
and (not request.documents)
}
number_fields = {
name: spec
for (name, spec) in request.fields.items()
if name in active
and spec.get("type") == "number"
and (spec.get("source_only") is True)
and (not spec.get("knowledge"))
and (not grounding.state_only(spec))
and (len(request.images) > 1 or request.documents)
}
selected = {**po_fields, **number_fields}
if not selected:
return encoded
originals = record.get("_original_images", ())
if len(originals) < len(request.images) or (
not request.documents and len(originals) != len(request.images)
):
raise ValueError(
"27B scalar detail must retain every original image in order"
)
(child, _, replacements, child_typed, child_eos) = super().encode(
request.model_copy(update={"fields": selected}),
identifier_view=True,
pdf_spelling_view=True,
image_views=lambda: originals,
image_view_key="27b-literal-scalars-all-content2-v1",
)
if child_typed is not None or child_eos != eos:
raise ValueError(
"27B scalar detail must retain the original scalar AR contract"
)
by_name = {task[0]: task for task in replacements}
if set(by_name) != set(selected) or len(replacements) != len(selected):
raise ValueError(
"27B scalar detail tasks differ from selected public fields"
)
updated = [by_name.get(task[0], task) for task in tasks]
salts = dict(record.get("_ar_task_salts", {}))
for name in selected:
record["fields"][name] = child["fields"][name]
record["_ar_constraints"][name] = child["_ar_constraints"][name]
salts[name] = child["_cache_salt"]
record["_meta"]["model_document_scalar_view"] = {
"version": "27b-literal-scalars-all-content2-v1",
"fields": list(selected),
"po_fields": list(po_fields),
"number_fields": list(number_fields),
"image_count": len(originals),
}
record.update(
_ar_tasks=updated,
_ar_task_salts=salts,
_ar_prefix=[],
_ar_group_prefixes=[],
)
record["_ar_media_groups"] = media_warmup_groups(
updated,
default_salt=record["_cache_salt"],
task_salts=salts,
annotations=record["_meta"]["extraction_contract"]["fields"],
)
return (record, forecasts, updated, typed, eos)
def _encode_request(self, request: PredictRequest) -> EncodedRequest:
encoded = super()._encode_request(request)
if (
type(self) is Clairic27BModel
and self.settings.model_id == "clairic-27b"
and (self.max_concurrency == self.settings.max_concurrency == 1)
and bool(
getattr(request, "images", ()) or getattr(request, "documents", ())
)
):
encoded[0]["_first_item_list_gating"] = True
fallback = encoded[0].get("_identifier_spelling_fallback")
if fallback is not None:
fallback["_first_item_list_gating"] = True
encoded = media_state_prompt.prepare(
request,
encoded,
tokenizer=self.tokenizer,
max_length=self.settings.max_length,
media_groups=media_warmup_groups,
)
return encoded
async def _ar_all(
self, record: dict[str, Any], eos: int
) -> dict[str, dict[str, Any]]:
caller_record = record
if len(record["_ar_tasks"]) < 64 and (not record.get("_media_payload")):
record = {**record, "_ar_prefix": None, "_ar_group_prefixes": []}
if (
type(self) is Clairic27BModel
and self.settings.model_id == "clairic-27b"
and (self.max_concurrency == self.settings.max_concurrency == 1)
and text_list_completion.eligible(record, record["_ar_tasks"])
):
predictions = await text_ar.text_ar_all(self, record, eos)
if record is not caller_record:
caller_record["_unexecuted_list_items"] = record[
"_unexecuted_list_items"
]
return predictions
return await super()._ar_all(record, eos)
|