File size: 11,546 Bytes
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 | """CLI wiring tests for --prompt-wording (argparse choices/env default, and
plumbing into SGLangBackend + create_app). uvicorn.run is monkeypatched so
main() never actually binds a socket; SGLangBackend/create_app/NativeTokenizer
are monkeypatched to capture their constructor arguments instead of talking to
a real engine or loading real tokenizer files (transformers is an optional,
not-installed-here dependency for this default venv)."""
import sys
import pytest
import jev_adapter.__main__ as main_module
import jev_adapter.native_tokenizer as native_tokenizer_module
import jev_adapter.server as server_module
import jev_adapter.sglang as sglang_module
class Capture:
def __init__(self):
self.backend_kwargs = None
self.app_kwargs = None
def fake_backend(self, *args, **kwargs):
self.backend_kwargs = kwargs
return object()
def fake_create_app(self, backend, **kwargs):
self.app_kwargs = kwargs
return object()
@pytest.fixture
def capture(monkeypatch):
capture = Capture()
monkeypatch.setattr(sglang_module, "SGLangBackend", capture.fake_backend)
monkeypatch.setattr(server_module, "create_app", capture.fake_create_app)
monkeypatch.setattr("uvicorn.run", lambda app, **kwargs: None)
return capture
def run_main(monkeypatch, argv, env=None):
monkeypatch.setattr(sys, "argv", ["jev-adapter", *argv])
for key in (
"JEV_PROMPT_WORDING",
"JEV_DEFAULT_TEMPERATURE",
"JEV_NATIVE_SYSTEM_PROMPT",
):
monkeypatch.delenv(key, raising=False)
for key, value in (env or {}).items():
monkeypatch.setenv(key, value)
main_module.main()
def test_prompt_wording_defaults_to_served(monkeypatch, capture):
run_main(monkeypatch, ["--model", "decision-model"])
assert capture.backend_kwargs["prompt_wording"] == "served"
assert capture.backend_kwargs["native_system_prompt"] is None
assert capture.app_kwargs["prompt_wording"] == "served"
def test_prompt_wording_flag_selects_native(monkeypatch, capture):
run_main(monkeypatch, ["--model", "decision-model", "--prompt-wording", "native"])
assert capture.backend_kwargs["prompt_wording"] == "native"
assert capture.app_kwargs["prompt_wording"] == "native"
def test_prompt_wording_env_var_sets_the_default(monkeypatch, capture):
run_main(
monkeypatch,
["--model", "decision-model"],
env={"JEV_PROMPT_WORDING": "native"},
)
assert capture.backend_kwargs["prompt_wording"] == "native"
assert capture.app_kwargs["prompt_wording"] == "native"
def test_explicit_flag_overrides_env_var(monkeypatch, capture):
run_main(
monkeypatch,
["--model", "decision-model", "--prompt-wording", "served"],
env={"JEV_PROMPT_WORDING": "native"},
)
assert capture.backend_kwargs["prompt_wording"] == "served"
def test_invalid_prompt_wording_choice_rejected(monkeypatch, capture):
monkeypatch.setattr(sys, "argv", ["jev-adapter", "--model", "m", "--prompt-wording", "bogus"])
with pytest.raises(SystemExit):
main_module.main()
def test_native_wording_without_tokenizer_model_serves_text_only_and_warns(
monkeypatch, capture, caplog
):
with caplog.at_level("WARNING"):
run_main(monkeypatch, ["--model", "decision-model", "--prompt-wording", "native"])
assert capture.backend_kwargs["native_system_prompt"] is None
assert any("native" in record.message for record in caplog.records)
def test_native_wording_with_tokenizer_model_extracts_the_system_prompt(
monkeypatch, capture
):
calls = []
class FakeNativeTokenizer:
@classmethod
def from_pretrained(cls, model, revision):
calls.append(("from_pretrained", model, revision))
return object()
@classmethod
def native_default_system_prompt(cls, model, revision):
calls.append(("native_default_system_prompt", model, revision))
return "Default system text."
monkeypatch.setattr(native_tokenizer_module, "NativeTokenizer", FakeNativeTokenizer)
run_main(
monkeypatch,
[
"--model",
"decision-model",
"--prompt-wording",
"native",
"--tokenizer-model",
"mistralai/Ministral-3-8B-Instruct-2512-BF16",
"--tokenizer-revision",
"f" * 40,
],
)
assert calls == [
("from_pretrained", "mistralai/Ministral-3-8B-Instruct-2512-BF16", "f" * 40),
(
"native_default_system_prompt",
"mistralai/Ministral-3-8B-Instruct-2512-BF16",
"f" * 40,
),
]
assert capture.backend_kwargs["native_system_prompt"] == "Default system text."
def test_served_wording_with_tokenizer_model_never_extracts_a_system_prompt(
monkeypatch, capture
):
calls = []
class FakeNativeTokenizer:
@classmethod
def from_pretrained(cls, model, revision):
calls.append("from_pretrained")
return object()
@classmethod
def native_default_system_prompt(cls, model, revision):
calls.append("native_default_system_prompt")
return "should not be reached"
monkeypatch.setattr(native_tokenizer_module, "NativeTokenizer", FakeNativeTokenizer)
run_main(
monkeypatch,
[
"--model",
"decision-model",
"--tokenizer-model",
"org/model",
"--tokenizer-revision",
"a" * 40,
],
)
assert calls == ["from_pretrained"]
assert capture.backend_kwargs["prompt_wording"] == "served"
assert capture.backend_kwargs["native_system_prompt"] is None
def test_native_system_prompt_defaults_to_auto(monkeypatch, capture):
calls = []
class FakeNativeTokenizer:
@classmethod
def from_pretrained(cls, model, revision):
return object()
@classmethod
def native_default_system_prompt(cls, model, revision):
calls.append((model, revision))
return "Default system text."
monkeypatch.setattr(native_tokenizer_module, "NativeTokenizer", FakeNativeTokenizer)
run_main(
monkeypatch,
[
"--model",
"decision-model",
"--prompt-wording",
"native",
"--tokenizer-model",
"org/model",
"--tokenizer-revision",
"a" * 40,
],
)
assert calls == [("org/model", "a" * 40)]
assert capture.backend_kwargs["native_system_prompt"] == "Default system text."
def test_native_system_prompt_none_skips_extraction_and_warning(
monkeypatch, capture, caplog
):
calls = []
class FakeNativeTokenizer:
@classmethod
def from_pretrained(cls, model, revision):
return object()
@classmethod
def native_default_system_prompt(cls, model, revision):
calls.append((model, revision))
return "should not be reached"
monkeypatch.setattr(native_tokenizer_module, "NativeTokenizer", FakeNativeTokenizer)
with caplog.at_level("WARNING"):
run_main(
monkeypatch,
[
"--model",
"decision-model",
"--prompt-wording",
"native",
"--tokenizer-model",
"org/model",
"--tokenizer-revision",
"a" * 40,
"--native-system-prompt",
"none",
],
)
assert calls == []
assert capture.backend_kwargs["native_system_prompt"] is None
assert not any("native" in record.message for record in caplog.records)
def test_native_system_prompt_none_without_tokenizer_never_warns(
monkeypatch, capture, caplog
):
with caplog.at_level("WARNING"):
run_main(
monkeypatch,
[
"--model",
"decision-model",
"--prompt-wording",
"native",
"--native-system-prompt",
"none",
],
)
assert capture.backend_kwargs["native_system_prompt"] is None
assert caplog.records == []
def test_native_system_prompt_env_var_sets_the_default(monkeypatch, capture, caplog):
with caplog.at_level("WARNING"):
run_main(
monkeypatch,
["--model", "decision-model", "--prompt-wording", "native"],
env={"JEV_NATIVE_SYSTEM_PROMPT": "none"},
)
assert capture.backend_kwargs["native_system_prompt"] is None
assert caplog.records == []
def test_native_system_prompt_explicit_flag_overrides_env_var(monkeypatch, capture):
calls = []
class FakeNativeTokenizer:
@classmethod
def from_pretrained(cls, model, revision):
return object()
@classmethod
def native_default_system_prompt(cls, model, revision):
calls.append((model, revision))
return "Default system text."
monkeypatch.setattr(native_tokenizer_module, "NativeTokenizer", FakeNativeTokenizer)
run_main(
monkeypatch,
[
"--model",
"decision-model",
"--prompt-wording",
"native",
"--tokenizer-model",
"org/model",
"--tokenizer-revision",
"a" * 40,
"--native-system-prompt",
"auto",
],
env={"JEV_NATIVE_SYSTEM_PROMPT": "none"},
)
assert calls == [("org/model", "a" * 40)]
assert capture.backend_kwargs["native_system_prompt"] == "Default system text."
def test_native_system_prompt_none_has_no_effect_on_served_wording(
monkeypatch, capture
):
calls = []
class FakeNativeTokenizer:
@classmethod
def from_pretrained(cls, model, revision):
return object()
@classmethod
def native_default_system_prompt(cls, model, revision):
calls.append((model, revision))
return "should not be reached"
monkeypatch.setattr(native_tokenizer_module, "NativeTokenizer", FakeNativeTokenizer)
run_main(
monkeypatch,
[
"--model",
"decision-model",
"--tokenizer-model",
"org/model",
"--tokenizer-revision",
"a" * 40,
"--native-system-prompt",
"auto",
],
)
assert calls == []
assert capture.backend_kwargs["prompt_wording"] == "served"
assert capture.backend_kwargs["native_system_prompt"] is None
def test_invalid_native_system_prompt_choice_rejected(monkeypatch, capture):
monkeypatch.setattr(
sys,
"argv",
["jev-adapter", "--model", "m", "--native-system-prompt", "bogus"],
)
with pytest.raises(SystemExit):
main_module.main()
def test_native_system_prompt_startup_log_line(monkeypatch, capture, caplog):
with caplog.at_level("INFO"):
run_main(
monkeypatch,
[
"--model",
"decision-model",
"--prompt-wording",
"native",
"--native-system-prompt",
"none",
],
)
info_records = [r for r in caplog.records if r.levelname == "INFO"]
assert len(info_records) == 1
message = info_records[0].getMessage()
assert "prompt_wording=native" in message
assert "native_system_prompt=none" in message
|