Text Generation
Transformers
Safetensors
mistral3
image-text-to-text
decision-model
typed-decisions
jev
jevbench
calibration
decode-free
multilingual
vision-language
conversational
Instructions to use StandardThinking/StandardOne-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use StandardThinking/StandardOne-3B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="StandardThinking/StandardOne-3B") 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("StandardThinking/StandardOne-3B") model = AutoModelForMultimodalLM.from_pretrained("StandardThinking/StandardOne-3B", 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 StandardThinking/StandardOne-3B with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "StandardThinking/StandardOne-3B" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "StandardThinking/StandardOne-3B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/StandardThinking/StandardOne-3B
- SGLang
How to use StandardThinking/StandardOne-3B 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 "StandardThinking/StandardOne-3B" \ --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": "StandardThinking/StandardOne-3B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'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 "StandardThinking/StandardOne-3B" \ --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": "StandardThinking/StandardOne-3B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use StandardThinking/StandardOne-3B with Docker Model Runner:
docker model run hf.co/StandardThinking/StandardOne-3B
File size: 17,075 Bytes
ea84b46 3c08bbf ea84b46 3c08bbf ea84b46 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 | import asyncio
import math
import unittest
from fastapi.testclient import TestClient
from jev_adapter.__main__ import positive_temperature, temperature_map
from jev_adapter.backend import AdapterError, ScoringResult
from jev_adapter.protocol import SystemOneRequest, probabilities_from_logprobs
from jev_adapter.server import create_app
from jev_adapter.service import SystemOneService
def payload():
return {
"model": "test-model",
"state": {"text": "결제가 두 번 되었어요. 환불해 주세요."},
"questions": {
"route": {
"type": "choice",
"instructions": "담당 부서",
"criteria": {"billing": "청구", "technical": "기술"},
},
"refund": {"type": "noul", "instructions": "환불 요청인가?"},
"urgency": {
"type": "score",
"instructions": "긴급도",
"criteria": ["낮음", "중간", "높음"],
},
},
}
class FakeBackend:
model = "test-model"
def __init__(self, delay=0, failure=False):
self.calls = []
self.active = self.peak = self.cancelled = 0
self.started = self.closed = False
self.delay, self.failure = delay, failure
async def start(self):
self.started = True
async def close(self):
self.closed = True
def labels(self, count):
return tuple(chr(65 + i) for i in range(count)), tuple(range(65, 65 + count))
async def evaluate(self, prompt, images, labels, token_ids, assistant_prefix):
self.calls.append((prompt, images, labels, token_ids, assistant_prefix))
self.active += 1
self.peak = max(self.peak, self.active)
try:
if self.failure:
raise AdapterError(
"backend_unavailable", "Engine unavailable.", status=502
)
await asyncio.sleep(self.delay)
return ScoringResult(
tuple(math.log(0.8 if i == 0 else 0.1) for i in range(len(labels))), 20
)
except asyncio.CancelledError:
self.cancelled += 1
raise
finally:
self.active -= 1
class TestService(unittest.IsolatedAsyncioTestCase):
async def test_mixed_questions_keep_images_and_build_typed_response(self):
backend = FakeBackend()
body = payload()
body["images"] = ["data:image/png;base64,aGVsbG8="]
result = await SystemOneService(backend).score(
SystemOneRequest.model_validate(body)
)
self.assertEqual(result["answers"]["route"]["choice"], "billing")
self.assertAlmostEqual(result["answers"]["refund"]["noul"], 8 / 9)
self.assertAlmostEqual(result["answers"]["urgency"]["score"], 0.3)
self.assertEqual(result["usage"], {"input_tokens": 60, "output_tokens": 0})
self.assertEqual(result["metadata"]["evaluations"], 3)
for call in backend.calls:
self.assertEqual(call[1], body["images"])
self.assertNotIn("담당 부서", backend.calls[1][0])
async def test_rotations_preserve_canonical_option_mapping(self):
body = payload()
body["questions"] = {"route": body["questions"]["route"]}
body["options"] = {"permutations": 2, "return_logprobs": True}
result = await SystemOneService(FakeBackend()).score(
SystemOneRequest.model_validate(body)
)
route = result["answers"]["route"]
self.assertAlmostEqual(route["probabilities"]["billing"], 0.5)
self.assertAlmostEqual(route["confidence"], 0)
self.assertEqual(len(route["logprobs"]), 2)
async def test_concurrency_limit_applies_across_requests(self):
backend = FakeBackend(delay=0.01)
service = SystemOneService(backend, max_concurrency=2)
await asyncio.gather(
*(
service.score(SystemOneRequest.model_validate(payload()))
for _ in range(3)
)
)
self.assertEqual(len(backend.calls), 9)
self.assertEqual(backend.peak, 2)
async def test_cancellation_closes_all_pending_work(self):
backend = FakeBackend(delay=10)
task = asyncio.create_task(
SystemOneService(backend, max_concurrency=2).score(
SystemOneRequest.model_validate(payload())
)
)
while backend.active != 2:
await asyncio.sleep(0)
task.cancel()
with self.assertRaises(asyncio.CancelledError):
await task
self.assertEqual(backend.cancelled, 2)
self.assertEqual(backend.active, 0)
self.assertEqual(len(backend.calls), 2)
async def test_wrong_model_rejected_before_inference(self):
backend = FakeBackend()
body = payload()
body["model"] = "wrong-model"
with self.assertRaises(AdapterError) as error:
await SystemOneService(backend).score(SystemOneRequest.model_validate(body))
self.assertEqual(error.exception.status, 404)
self.assertEqual(backend.calls, [])
class TestDefaultTemperature(unittest.IsolatedAsyncioTestCase):
"""A server-side default fills an omitted options.temperature only."""
vector = tuple(math.log(0.8 if i == 0 else 0.1) for i in range(3))
async def score(self, body, **service_options):
service = SystemOneService(FakeBackend(), **service_options)
return await service.score(SystemOneRequest.model_validate(body))
def route_probabilities(self, result):
return list(result["answers"]["urgency"]["probabilities"].values())
async def test_unset_default_keeps_request_default_of_one(self):
result = await self.score(payload())
self.assertEqual(result["metadata"]["temperature"], 1.0)
expected = probabilities_from_logprobs(self.vector, 1.0)
for actual, wanted in zip(self.route_probabilities(result), expected):
self.assertAlmostEqual(actual, wanted)
async def test_omitted_temperature_takes_server_default(self):
for body in (payload(), {**payload(), "options": {"permutations": 1}}):
result = await self.score(body, default_temperature=2.4)
self.assertEqual(result["metadata"]["temperature"], 2.4)
expected = probabilities_from_logprobs(self.vector, 2.4)
for actual, wanted in zip(self.route_probabilities(result), expected):
self.assertAlmostEqual(actual, wanted)
self.assertNotAlmostEqual(
self.route_probabilities(result)[0],
probabilities_from_logprobs(self.vector, 1.0)[0],
)
async def test_explicit_request_temperature_wins_over_default(self):
body = {**payload(), "options": {"temperature": 1.0}}
result = await self.score(body, default_temperature=2.4)
self.assertEqual(result["metadata"]["temperature"], 1.0)
expected = probabilities_from_logprobs(self.vector, 1.0)
for actual, wanted in zip(self.route_probabilities(result), expected):
self.assertAlmostEqual(actual, wanted)
body = {**payload(), "options": {"temperature": 3.0}}
result = await self.score(body, default_temperature=2.4)
self.assertEqual(result["metadata"]["temperature"], 3.0)
async def test_disabled_scaling_ignores_default(self):
body = {**payload(), "options": {"temperature_scaling": False}}
result = await self.score(body, default_temperature=2.4)
self.assertEqual(result["metadata"]["temperature"], 1.0)
expected = probabilities_from_logprobs(self.vector, 1.0)
for actual, wanted in zip(self.route_probabilities(result), expected):
self.assertAlmostEqual(actual, wanted)
async def test_default_does_not_mutate_the_request(self):
request = SystemOneRequest.model_validate(payload())
service = SystemOneService(FakeBackend(), default_temperature=2.4)
await service.score(request)
self.assertEqual(request.options.temperature, 1.0)
self.assertNotIn("temperature", request.options.model_fields_set)
self.assertEqual(service.effective_options(request).temperature, 2.4)
def test_invalid_default_temperature_rejected_at_construction(self):
for value in (0, -1.0, float("inf"), float("nan"), True, "2.4", None):
with self.assertRaises((ValueError, TypeError)):
SystemOneService(FakeBackend(), default_temperature=value)
self.assertEqual(
SystemOneService(FakeBackend(), default_temperature=2).default_temperature,
2.0,
)
async def test_temperature_by_type_applies_per_answer_type(self):
result = await self.score(
payload(), default_temperature=2.4, temperature_by_type={"score": 3.0, "noul": 1.5}
)
urgency = list(result["answers"]["urgency"]["probabilities"].values())
for actual, wanted in zip(urgency, probabilities_from_logprobs(self.vector, 3.0)):
self.assertAlmostEqual(actual, wanted)
route = list(result["answers"]["route"]["probabilities"].values())
for actual, wanted in zip(route, probabilities_from_logprobs(self.vector[:2], 2.4)):
self.assertAlmostEqual(actual, wanted)
self.assertEqual(result["metadata"]["temperature"], 2.4)
self.assertEqual(result["metadata"]["temperature_by_type"], {"score": 3.0, "noul": 1.5})
async def test_explicit_request_temperature_wins_over_temperature_by_type(self):
body = {**payload(), "options": {"temperature": 1.0}}
result = await self.score(body, temperature_by_type={"score": 3.0})
urgency = list(result["answers"]["urgency"]["probabilities"].values())
for actual, wanted in zip(urgency, probabilities_from_logprobs(self.vector, 1.0)):
self.assertAlmostEqual(actual, wanted)
self.assertNotIn("temperature_by_type", result["metadata"])
async def test_disabled_scaling_ignores_temperature_by_type(self):
body = {**payload(), "options": {"temperature_scaling": False}}
result = await self.score(body, temperature_by_type={"score": 3.0})
urgency = list(result["answers"]["urgency"]["probabilities"].values())
for actual, wanted in zip(urgency, probabilities_from_logprobs(self.vector, 1.0)):
self.assertAlmostEqual(actual, wanted)
self.assertNotIn("temperature_by_type", result["metadata"])
def test_invalid_temperature_by_type_rejected_at_construction(self):
for value in ({"judge": 1.2}, {"score": 0}, {"score": float("nan")}, {"noul": True}):
with self.assertRaises(ValueError):
SystemOneService(FakeBackend(), temperature_by_type=value)
self.assertEqual(SystemOneService(FakeBackend()).temperature_by_type, {})
def test_cli_temperature_map_type(self):
self.assertEqual(temperature_map("choice=1.6, noul=1.2,score=1.8"), {"choice": 1.6, "noul": 1.2, "score": 1.8})
self.assertEqual(temperature_map(""), {})
import argparse
for text in ("choice", "judge=1.2", "choice=0", "choice=1.2,choice=1.3", "noul=abc"):
with self.assertRaises(argparse.ArgumentTypeError):
temperature_map(text)
def test_cli_default_temperature_type(self):
self.assertEqual(positive_temperature("2.4"), 2.4)
self.assertEqual(positive_temperature("1"), 1.0)
import argparse
for text in ("0", "-1", "inf", "nan", "abc", ""):
with self.assertRaises(argparse.ArgumentTypeError):
positive_temperature(text)
class TestPromptWording(unittest.IsolatedAsyncioTestCase):
"""--prompt-wording is a server-wide setting (CLI/env, not per-request);
the default keeps today's served text unchanged."""
async def test_default_served_wording_is_unchanged(self):
backend = FakeBackend()
body = payload()
body["questions"] = {"route": body["questions"]["route"]}
await SystemOneService(backend).score(SystemOneRequest.model_validate(body))
prompt = backend.calls[0][0]
self.assertTrue(prompt.startswith("Context:\n"))
async def test_native_wording_is_plumbed_into_build_prompt(self):
backend = FakeBackend()
body = payload()
body["questions"] = {"route": body["questions"]["route"]}
service = SystemOneService(backend, prompt_wording="native")
await service.score(SystemOneRequest.model_validate(body))
prompt = backend.calls[0][0]
self.assertTrue(prompt.startswith("Read the state and question."))
self.assertIn("\n\nState:\n", prompt)
self.assertNotIn("Context:\n", prompt)
async def test_native_wording_requires_canonical_az_labels(self):
# FakeBackend.labels() already returns canonical A, B, C, ... so this
# documents the happy path; build_prompt itself enforces the
# requirement (see TestNativePromptWording in test_protocol.py) when a
# backend's labels are not canonical.
backend = FakeBackend()
labels, _ = backend.labels(3)
self.assertEqual(labels, ("A", "B", "C"))
def test_invalid_prompt_wording_rejected_at_construction(self):
for value in ("", "SERVED", "native ", None, 1):
with self.assertRaises(ValueError):
SystemOneService(FakeBackend(), prompt_wording=value)
class TestHTTP(unittest.TestCase):
def test_standalone_app_lifecycle_schema_and_alias(self):
backend = FakeBackend()
with TestClient(create_app(backend)) as client:
self.assertTrue(backend.started)
self.assertEqual(client.get("/health").json(), {"status": "ok"})
self.assertEqual(
client.get("/v1/models").json()["data"][0]["id"], backend.model
)
body = payload()
body["model"] = "jev-latest"
response = client.post("/v1/systemone", json=body)
self.assertEqual(response.status_code, 200)
self.assertEqual(response.json()["model"], backend.model)
self.assertEqual(
set(response.json()["answers"]), {"route", "refund", "urgency"}
)
bad = client.post("/v1/systemone", json={"model": backend.model})
self.assertEqual(bad.status_code, 422)
self.assertEqual(bad.json()["error"]["code"], "invalid_request")
self.assertTrue(backend.closed)
def test_app_default_temperature_applies_when_request_omits_it(self):
with TestClient(create_app(FakeBackend(), default_temperature=2.4)) as client:
response = client.post("/v1/systemone", json=payload())
self.assertEqual(response.status_code, 200)
self.assertEqual(response.json()["metadata"]["temperature"], 2.4)
body = {**payload(), "options": {"temperature": 1.5}}
response = client.post("/v1/systemone", json=body)
self.assertEqual(response.json()["metadata"]["temperature"], 1.5)
with self.assertRaises(ValueError):
create_app(FakeBackend(), default_temperature=0)
def test_app_prompt_wording_native_reaches_the_backend_over_http(self):
backend = FakeBackend()
with TestClient(create_app(backend, prompt_wording="native")) as client:
body = payload()
body["questions"] = {"route": body["questions"]["route"]}
response = client.post("/v1/systemone", json=body)
self.assertEqual(response.status_code, 200)
self.assertTrue(backend.calls[0][0].startswith("Read the state and question."))
def test_auth_guards_inference_and_model_discovery(self):
backend = FakeBackend()
with TestClient(create_app(backend, api_key="test-secret")) as client:
self.assertEqual(client.get("/v1/models").status_code, 401)
self.assertEqual(
client.post("/v1/systemone", json=payload()).status_code, 401
)
self.assertEqual(
client.post(
"/v1/systemone",
json=payload(),
headers={b"Authorization": b"Bearer caf\xe9"},
).status_code,
401,
)
self.assertEqual(backend.calls, [])
response = client.post(
"/v1/systemone",
json=payload(),
headers={"Authorization": "Bearer test-secret"},
)
self.assertEqual(response.status_code, 200)
def test_upstream_failure_is_not_returned_as_probabilities(self):
with TestClient(create_app(FakeBackend(failure=True))) as client:
response = client.post("/v1/systemone", json=payload())
self.assertEqual(response.status_code, 502)
self.assertEqual(response.json()["error"]["code"], "backend_unavailable")
self.assertNotIn("answers", response.json())
|