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: 9,206 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 | from __future__ import annotations
from typing import TYPE_CHECKING
from . import batch_completion as shared
from . import ordinary_list_completion, schema
if TYPE_CHECKING:
from collections.abc import Callable
from typing import Any
from .backend import ARTask
from transformers import PreTrainedTokenizerBase
FRONTIER = 8
def _object_lists(
record: dict[str, Any], tasks: list[ARTask]
) -> dict[str, tuple[int, ...]]:
if record.get("_null_recovery"):
return {}
fields = record["fields"]
annotations = (
record.get("_meta", {}).get("extraction_contract", {}).get("fields", {})
)
names = [name for (name, _, _) in tasks]
if len(set(names)) != len(names):
raise ValueError("AR task names must be distinct")
indices = {name: index for (index, name) in enumerate(names)}
claimed: set[str]
groups: dict[str, tuple[int, ...]]
(claimed, groups) = (set(), {})
forbidden = {"_physical_row_scope", "_image_row_inventory", "_image_row_proposal"}
plans = (
record.get("_financial_line_plans", {}),
record.get("_list_source_bounds", {}),
)
for parent, annotation in annotations.items():
spec = fields.get(parent, {})
queries = annotation.get("query_fields")
item = spec.get("items")
if (
spec.get("type") != "list"
or not isinstance(item, dict)
or item.get("type") != "object"
or spec.get("knowledge")
or (set(annotation) != {"query_fields"})
or (not isinstance(queries, list))
or (len(queries) <= FRONTIER)
or any((spec.get(key) for key in forbidden))
or (parent in indices)
or any((parent in plan for plan in plans))
):
continue
valid = True
for ordinal, name in enumerate(queries, 1):
child = fields.get(name, {})
ownership = annotations.get(name, {})
if (
name != schema.list_item_name(parent, ordinal)
or name not in indices
or name in claimed
or (ownership != {"parent": parent, "index": ordinal})
or (child.get("type") != "object")
or (child.get("nullable", True) is not True)
or any((child.get(key) for key in forbidden))
or (name in record.get("_source_row_gates", ()))
):
valid = False
break
if "enum" in child and None not in child["enum"]:
valid = False
break
if valid:
groups[parent] = tuple((indices[name] for name in queries))
claimed.update(queries)
return groups
def _groups(record: dict[str, Any], tasks: list[ARTask]) -> dict[str, tuple[int, ...]]:
if (
record.get("_null_recovery")
or record.get("_meta", {}).get("construction") != schema.EVIDENCE_CONSTRUCTION
):
return {}
groups = _object_lists(record, tasks)
if (
not record.get("_media_payload")
and (not record.get("_original_images"))
and all((media is None for (_, _, media) in tasks))
):
groups.update(
{
parent: group
for (parent, group) in shared._ordinary_lists(
record, tasks, schema.list_item_name
).items()
if len(group) > FRONTIER
}
)
return groups
def eligible(record: dict[str, Any], tasks: list[ARTask]) -> bool:
return bool(_groups(record, tasks))
def completed_null(
result: dict[str, Any], eos: int, tokenizer: PreTrainedTokenizerBase
) -> bool:
try:
return shared._completed_null(result, eos, tokenizer)
except (KeyError, TypeError, ValueError, OverflowError):
return False
async def generate(
engine: Any,
record: dict[str, Any],
tasks: list[ARTask],
eos: int,
tokenizer: PreTrainedTokenizerBase,
*,
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]]:
if eligible(record, tasks) and ordinary_list_completion.eligible(
record, tasks, kwargs
):
return await ordinary_list_completion.generate(
engine,
record,
tasks,
eos,
tokenizer,
initial_frontier=8,
next_frontier=32,
parse_result=parse_result,
normalize=normalize,
**kwargs,
)
media = kwargs.get("image_data")
if (
set(kwargs) - shared._BATCHED_KEYS - shared._SHARED_KEYS
or not eligible(record, tasks)
or (not isinstance(media, list))
or (len(media) != len(tasks))
or any(
(
actual is not task[2]
and (
not (task[2] is None and isinstance(actual, list) and (not actual))
)
for (actual, task) in zip(media, tasks, strict=True)
)
)
):
return await shared.generate(
engine,
record,
tasks,
eos,
tokenizer,
parse_result=parse_result,
normalize=normalize,
**kwargs,
)
groups = _groups(record, tasks)
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("List frontier requires the original distinct AR batch")
shared._select_kwargs(kwargs, list(range(size)), size)
deferred = {index for group in groups.values() for index in group[FRONTIER:]}
results: list[Any] = [None] * size
lengths = {name: len(ids) for (name, ids, _media) in tasks}
unexecuted: dict[str, int] = {}
async def submit(indices: list[int]) -> None:
if not indices:
return
selected = shared._select_kwargs(kwargs, indices, size)
subset = [tasks[index] for index in indices]
received = await shared.generate(
engine,
record,
subset,
eos,
tokenizer,
parse_result=parse_result,
normalize=normalize,
**selected,
)
accounted = record.get("_unexecuted_list_items", {})
if any(
(
name not in lengths or length != lengths[name]
for (name, length) in accounted.items()
)
):
raise RuntimeError("Shared scheduler changed original omission accounting")
if len(received) != len(indices):
raise RuntimeError(
"Native completion differs from its list-frontier submission"
)
for index, result, rid in zip(indices, received, selected["rid"], strict=True):
name = tasks[index][0]
if name in accounted:
if (
not result.get("_clair_unused_row_proposal")
or result.get("meta_info", {}).get("completion_tokens") != 0
):
raise RuntimeError(
"Shared scheduler omission lacks its unexecuted result slot"
)
elif result.get("meta_info", {}).get("id") != rid:
raise RuntimeError(
"Native completion differs from its list-frontier submission"
)
results[index] = result
for name, length in accounted.items():
if name in unexecuted and unexecuted[name] != length:
raise RuntimeError(
"Schedulers disagree on original omitted prompt length"
)
unexecuted[name] = length
await submit([index for index in range(size) if index not in deferred])
continued: list[int] = []
for group in groups.values():
if any(
(
completed_null(results[index], eos, tokenizer)
for index in group[:FRONTIER]
)
):
for index in group[FRONTIER:]:
(name, ids, _) = tasks[index]
if name in unexecuted:
raise RuntimeError("List frontier attempted to omit a slot twice")
unexecuted[name] = len(ids)
results[index] = {
"_clair_unused_row_proposal": True,
"_text_frontier_unexecuted": True,
"meta_info": {"completion_tokens": 0},
}
else:
continued.extend(group[FRONTIER:])
await submit(sorted(continued))
if any((result is None for result in results)):
raise RuntimeError("List frontier left an unaccounted result slot")
record["_unexecuted_list_items"] = unexecuted
return results
|