Image-Text-to-Text
MLX
Safetensors
qwen3_5
clairic
cloudflare
systemone
structured-output
information-extraction
multimodal
custom-code
conversational
4-bit precision
Instructions to use mlx-community/clairic-flash-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use mlx-community/clairic-flash-4bit with MLX:
# Make sure mlx-vlm is installed # pip install --upgrade mlx-vlm from mlx_vlm import load, generate from mlx_vlm.prompt_utils import apply_chat_template from mlx_vlm.utils import load_config # Load the model model, processor = load("mlx-community/clairic-flash-4bit") config = load_config("mlx-community/clairic-flash-4bit") # Prepare input image = ["http://images.cocodataset.org/val2017/000000039769.jpg"] prompt = "Describe this image." # Apply chat template formatted_prompt = apply_chat_template( processor, config, prompt, num_images=1 ) # Generate output output = generate(model, processor, formatted_prompt, image) print(output) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Pi
How to use mlx-community/clairic-flash-4bit with Pi:
Start the MLX server
# Install MLX LM: uv tool install mlx-lm # Start a local OpenAI-compatible server: mlx_lm.server --model "mlx-community/clairic-flash-4bit"
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "mlx-lm": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "mlx-community/clairic-flash-4bit" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Hermes Agent
How to use mlx-community/clairic-flash-4bit with Hermes Agent:
Start the MLX server
# Install MLX LM: uv tool install mlx-lm # Start a local OpenAI-compatible server: mlx_lm.server --model "mlx-community/clairic-flash-4bit"
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default mlx-community/clairic-flash-4bit
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use mlx-community/clairic-flash-4bit with OpenClaw:
Start the MLX server
# Install MLX LM: uv tool install mlx-lm # Start a local OpenAI-compatible server: mlx_lm.server --model "mlx-community/clairic-flash-4bit"
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "mlx-community/clairic-flash-4bit" \ --custom-provider-id mlx-lm \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
File size: 12,291 Bytes
ae97c5d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 | from __future__ import annotations
import hashlib
import json
import re
import unicodedata
from collections import defaultdict
from collections.abc import Iterable, Sequence
from copy import deepcopy
from decimal import Decimal, localcontext
from typing import Any
from . import ar, grounding
MAX_SOURCE_CANDIDATES = 512
MAX_OPERANDS = 32
_SUM = re.compile("^\\s*(?:the\\s+)?sum\\s+of\\b", re.IGNORECASE)
_PROTECTED = re.compile(
"\\b(?:not|never|no|without|don't|avoid|unless|if|copy|copies|copied|printed|verbatim|literal|quoted|label|named|called)\\b",
re.IGNORECASE,
)
_CURRENCY_CODES = frozenset(
{
"USD",
"EUR",
"GBP",
"JPY",
"CNY",
"INR",
"AUD",
"CAD",
"CHF",
"HKD",
"NZD",
"SGD",
"SEK",
"NOK",
"DKK",
"KRW",
"BRL",
"MXN",
"ZAR",
}
)
_AMOUNT_PREFIX = re.compile(
"(?P<open>\\()?[ \\t]*(?P<before>[+-]?)[ \\t]*(?P<currency>\\$|[A-Z]{3})?[ \\t]*(?P<after>[+-]?)[ \\t]*$"
)
INSTRUCTION = "Select the printed numeric source occurrences that are operands of the requested sum. Follow the field instructions to identify the requested entities, rows, scope and inclusions. Return exactly one compact JSON array of candidate IDs in their source order. Include every matching occurrence once, even when different occurrences have equal amounts. Select the final amount for each requested record; do not also select its component amounts when those components are already included in that final amount. Exclude unrelated amounts, dates, identifiers and aggregate totals outside the requested scope. Do not calculate the sum, invent an amount, deduplicate equal source occurrences or reuse an ID. The original document is reproduced verbatim below; candidate contexts are source excerpts. Return null if no requested operands are present or the operand scope cannot be determined. A printed zero or negative amount is present. End immediately after the array or null."
def canonical(value: object) -> bytes:
return json.dumps(
value,
sort_keys=True,
ensure_ascii=False,
separators=(",", ":"),
allow_nan=False,
).encode()
def eligible(record: dict[str, Any], name: str, *, has_media: bool = False) -> bool:
spec = record["fields"][name]
annotation = (
record.get("_meta", {})
.get("extraction_contract", {})
.get("fields", {})
.get(name, {})
)
instructions = spec.get("instructions", "")
return (
not has_media
and (not record.get("images"))
and (not record.get("documents"))
and (not record.get("_media_payload"))
and (spec.get("type") == "number")
and (not spec.get("source_only"))
and (not spec.get("knowledge"))
and (not spec.get("_physical_row_scope"))
and (not annotation.get("parent"))
and (name not in record.get("_source_row_gates", ()))
and (not grounding.state_only(spec))
and (not grounding.document_only(spec))
and isinstance(instructions, str)
and bool(_SUM.search(instructions))
and (not _PROTECTED.search(instructions))
)
def _signed_readings(
source: str, start: int, end: int, values: Iterable[Decimal]
) -> set[Decimal]:
from .model import NUMERAL_RUN
numeral = NUMERAL_RUN.search(source, start, end)
if numeral is None:
return set()
left = numeral.start()
prefix_start = left
while prefix_start > max(0, left - 64):
char = source[prefix_start - 1]
if char in " \t(+-−" or unicodedata.category(char) == "Sc":
prefix_start -= 1
elif source[
max(0, prefix_start - 3) : prefix_start
].upper() in _CURRENCY_CODES and (
prefix_start == 3 or not source[prefix_start - 4].isalnum()
):
prefix_start -= 3
else:
break
prefix = source[prefix_start:left]
first = prefix_start + len(prefix) - len(prefix.lstrip())
if prefix.strip() in ("-", "−") and re.match(
"\\d{4}-\\d{2}-\\d{2}(?!\\d)", source[left:]
):
return {value.copy_abs() for value in values}
if (
first < left
and source[first] in "+-−"
and first
and source[first - 1].isalnum()
):
if prefix.strip() in ("+", "-", "−"):
return {value.copy_abs() for value in values}
return set()
normalized = "".join(
(
"$"
if unicodedata.category(char) == "Sc"
else "-"
if char == "−"
else char.upper()
for char in prefix
)
).lstrip()
match = _AMOUNT_PREFIX.fullmatch(normalized)
if match is None or (match["before"] and match["after"]):
return set()
explicit_sign = match["before"] or match["after"]
negative = explicit_sign == "-"
if match["open"]:
after = source[end:] if source[end - 1 : end] != ")" else source[end - 1 :]
closing = re.match("[ \\t]*\\)", after)
if closing is None or explicit_sign:
return set()
rest = after[closing.end() :].split("\n", 1)[0].strip(" \t.;,")
negative = bool(
match["currency"] or any((c in numeral[0] for c in ".,")) or (not rest)
)
return {
value.copy_abs().copy_negate() if negative else value.copy_abs()
for value in values
}
def _source_candidates(source: str) -> list[dict[str, Any]] | None:
readings = defaultdict(set)
for value, spans in grounding.numbers(source).items():
for span in spans:
readings[tuple(span)].add(value)
ordered = sorted(readings)
if not ordered or len(ordered) > MAX_SOURCE_CANDIDATES:
return None
previous_end = 0
candidates = []
for index, (start, end) in enumerate(ordered):
if start < previous_end or not 0 <= start < end <= len(source):
return None
previous_end = end
line_start = source.rfind("\n", 0, start) + 1
line_end = source.find("\n", end)
if line_end < 0:
line_end = len(source)
(context_start, context_end) = (
max(line_start, start - 64),
min(line_end, end + 64),
)
values = _signed_readings(source, start, end, readings[start, end])
candidates.append(
{
"id": f"N{index}",
"span": [start, end],
"printed": source[start:end],
"decimal": str(next(iter(values))) if len(values) == 1 else None,
"decimal_options": sorted((str(value) for value in values)),
"context": source[context_start:context_end],
}
)
return candidates
def _grammar(candidates: Sequence[dict[str, Any]]) -> str:
lines = ['root ::= "null" | "[" p0 "]"']
for index, item in enumerate(candidates):
literal = json.dumps(json.dumps(item["id"]))
rule = f"p{index} ::= {literal}"
if index + 1 < len(candidates):
rule += f' | {literal} "," p{index + 1} | p{index + 1}'
lines.append(rule)
return "\n".join(lines)
def prepare_task(
record: dict[str, Any], field_name: str, *, has_media: bool = False
) -> dict[str, Any] | None:
if not eligible(record, field_name, has_media=has_media):
return None
source = record["input"]
candidates = _source_candidates(source)
if not candidates:
return None
spec = deepcopy(record["fields"][field_name])
displayed = [
{"id": item["id"], "printed": item["printed"], "context": item["context"]}
for item in candidates
]
schema = {
"anyOf": [
{"type": "null"},
{
"type": "array",
"minItems": 1,
"maxItems": MAX_OPERANDS,
"uniqueItems": True,
"items": {
"type": "string",
"enum": [item["id"] for item in candidates],
},
},
]
}
prompt = (
INSTRUCTION
+ "\nField: "
+ field_name
+ "\nField instructions: "
+ spec.get("instructions", "")
+ "\nCandidates in source order: "
+ json.dumps(displayed, ensure_ascii=False, separators=(",", ":"))
+ "\n\nOriginal document:\n"
+ source
)
return {
"version": 2,
"kind": "source-occurrence-sum-v2",
"field": field_name,
"input_text": source,
"prompt": prompt,
"grammar": {"ebnf": _grammar(candidates)},
"json_schema": schema,
"candidates": candidates,
"field_spec": spec,
"max_operands": MAX_OPERANDS,
"source_sha256": hashlib.sha256(source.encode()).hexdigest(),
"field_spec_sha256": hashlib.sha256(canonical(spec)).hexdigest(),
"semantic_completeness_proven_by_grounding": False,
}
def validate_result(
task: dict[str, Any], raw_json: str, *, complete: bool = True
) -> dict[str, Any]:
result: dict[str, Any] = {
"valid": False,
"value": None,
"operands": [],
"decimal_sum": None,
"operand_count": 0,
"error": None,
"semantic_completeness_proven_by_grounding": False,
}
try:
if not complete:
raise ValueError("incomplete_generation")
source = task["input_text"]
if hashlib.sha256(source.encode()).hexdigest() != task["source_sha256"]:
raise ValueError("source_changed")
if (
hashlib.sha256(canonical(task["field_spec"])).hexdigest()
!= task["field_spec_sha256"]
):
raise ValueError("field_spec_changed")
ids = json.loads(
raw_json,
parse_constant=lambda _: (_ for _ in ()).throw(
ValueError("nonfinite_json")
),
)
if ids is None:
from jsonschema import Draft202012Validator
Draft202012Validator(ar.schema(task["field_spec"])).validate(None)
result.update(valid=True, error=None)
return result
if not isinstance(ids, list) or not ids or len(ids) > task["max_operands"]:
raise ValueError("invalid_operand_count")
if any((not isinstance(identifier, str) for identifier in ids)):
raise ValueError("operand_id_not_string")
if len(set(ids)) != len(ids):
raise ValueError("reused_source_occurrence")
source_candidates = _source_candidates(source)
if source_candidates != task["candidates"]:
raise ValueError("source_candidate_mapping_changed")
indexed = {
item["id"]: (index, item) for (index, item) in enumerate(source_candidates)
}
if any((identifier not in indexed for identifier in ids)):
raise ValueError("unknown_source_occurrence")
indices = [indexed[identifier][0] for identifier in ids]
if indices != sorted(indices):
raise ValueError("source_order_changed")
operands = [deepcopy(indexed[identifier][1]) for identifier in ids]
if any((item["decimal"] is None for item in operands)):
raise ValueError("ambiguous_operand_spelling")
amounts = [Decimal(item["decimal"]) for item in operands]
highest = max((amount.adjusted() for amount in amounts))
lowest: int = min((amount.as_tuple().exponent for amount in amounts))
with localcontext() as context:
context.prec = max(
context.prec, highest - lowest + len(str(len(amounts))) + 3
)
total = sum(amounts, Decimal(0))
value = ar.normalize_number(total, task["field_spec"])
from jsonschema import Draft202012Validator
Draft202012Validator(ar.schema(task["field_spec"])).validate(value)
result.update(
valid=True,
value=value,
operands=operands,
decimal_sum=str(total),
operand_count=len(operands),
)
except Exception as error:
result["error"] = str(error) or type(error).__name__
return result
|