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: 5,655 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 | from __future__ import annotations
from typing import TYPE_CHECKING, Any
from . import batch_completion as shared
from . import schema
if TYPE_CHECKING:
from collections.abc import Callable
from transformers import PreTrainedTokenizerBase
def eligible(
record: dict[str, Any],
tasks: list[tuple[str, list[int], Any]],
kwargs: dict[str, Any],
) -> bool:
media = kwargs.get("image_data")
object_lists = any(
(
spec.get("type") == "list"
and isinstance(spec.get("items"), dict)
and (spec["items"].get("type") == "object")
for spec in record.get("fields", {}).values()
)
)
guarded = (
not record.get("_null_recovery")
and (not object_lists)
and (
record.get("_meta", {}).get("construction") == schema.EVIDENCE_CONSTRUCTION
)
and (not record.get("_media_payload"))
and (not record.get("_original_images"))
and all((payload is None for (_, _, payload) in tasks))
and isinstance(media, list)
and (len(media) == len(tasks))
and all((payload is None for payload in media))
and (
not set(kwargs)
- shared._BATCHED_KEYS
- shared._SHARED_KEYS
- {"parse_result", "normalize"}
)
)
return guarded and bool(
shared._ordinary_lists(record, tasks, schema.list_item_name)
)
async def generate(
engine: Any,
record: dict[str, Any],
tasks: list[tuple[str, list[int], Any]],
eos: int,
tokenizer: PreTrainedTokenizerBase,
*,
initial_frontier: int = 0,
next_frontier: int = 0,
parse_result: Callable[..., dict[str, Any]] | None = None,
normalize: Callable[
[dict[str, Any], dict[str, dict[str, Any]]], dict[str, dict[str, Any]]
]
| None = None,
**kwargs: Any,
) -> list[dict[str, Any]]:
enabled = initial_frontier != 0
if enabled and (
type(initial_frontier) is not int
or type(next_frontier) is not int
or initial_frontier < 1
or (next_frontier <= initial_frontier)
):
raise ValueError(
"Ordinary list scheduling requires positive increasing integer endpoints"
)
groups = (
shared._ordinary_lists(record, tasks, schema.list_item_name)
if enabled and eligible(record, tasks, kwargs)
else {}
)
if not groups:
return await shared.generate(
engine,
record,
tasks,
eos,
tokenizer,
parse_result=parse_result,
normalize=normalize,
**kwargs,
)
size = len(tasks)
rids = kwargs.get("rid")
if (
not isinstance(rids, list)
or len(rids) != size
or len(set(rids)) != size
or (not all((isinstance(rid, str) and rid for rid in rids)))
or (kwargs.get("input_ids") != [ids for (_, ids, _) in tasks])
):
raise ValueError(
"Ordinary list scheduling requires original distinct task bindings"
)
shared._select_kwargs(kwargs, list(range(size)), size)
results: list[Any] = [None] * size
unexecuted: dict[str, int] = {}
async def submit(indices: list[int]) -> None:
if not indices:
return
selected = shared._select_kwargs(kwargs, indices, size)
received = await shared._generate(
engine,
record,
[tasks[index] for index in indices],
eos,
tokenizer,
**selected,
)
if len(received) != len(indices):
raise RuntimeError("Ordinary list scheduling received wrong batch size")
for index, result, rid in zip(indices, received, selected["rid"], strict=True):
if result.get("meta_info", {}).get("id") != rid:
raise RuntimeError(
"Ordinary list scheduling received wrong task identity"
)
results[index] = result
tails = {index for group in groups.values() for index in group[initial_frontier:]}
await submit([index for index in range(size) if index not in tails])
active = dict(groups)
(start, end, following) = (0, initial_frontier, next_frontier)
while active:
continued: list[int] = []
for parent, group in list(active.items()):
stopped = False
for index in group[start:end]:
try:
if shared._completed_null(results[index], eos, tokenizer):
stopped = True
break
except (KeyError, TypeError, ValueError, OverflowError):
continue
if stopped:
for index in group[end:]:
(name, ids, _) = tasks[index]
unexecuted[name] = len(ids)
results[index] = {
"_clair_unused_row_proposal": True,
"_ordinary_list_unexecuted": True,
"meta_info": {"completion_tokens": 0},
}
del active[parent]
elif end >= len(group):
del active[parent]
else:
continued.extend(group[end:following])
if continued:
await submit(sorted(continued))
(start, end, following) = (end, following, following * 2)
if any((result is None for result in results)):
raise RuntimeError("Ordinary list scheduling left an unaccounted result")
record["_unexecuted_list_items"] = unexecuted
return results
|