| 1 |
"""MLX serve launch artifact generation.""" |
| 2 |
|
| 3 |
from __future__ import annotations |
| 4 |
|
| 5 |
from pathlib import Path |
| 6 |
|
| 7 |
import pytest |
| 8 |
|
| 9 |
from dlm.base_models import BASE_MODELS |
| 10 |
from dlm.export.errors import ExportError, TargetSmokeError |
| 11 |
from dlm.export.manifest import load_export_manifest |
| 12 |
from dlm.export.targets.base import TargetResult |
| 13 |
from dlm.export.targets.mlx_serve import ( |
| 14 |
LAUNCH_SCRIPT_FILENAME, |
| 15 |
MLX_SERVE_TARGET, |
| 16 |
_quote_script_arg, |
| 17 |
_require_prepared_int, |
| 18 |
_require_prepared_path, |
| 19 |
_require_prepared_str, |
| 20 |
_version_from_dir_name, |
| 21 |
finalize_mlx_serve_export, |
| 22 |
prepare_mlx_serve_export, |
| 23 |
) |
| 24 |
from dlm.store.manifest import Manifest, load_manifest, save_manifest |
| 25 |
from dlm.store.paths import for_dlm |
| 26 |
|
| 27 |
_SPEC = BASE_MODELS["smollm2-135m"] |
| 28 |
|
| 29 |
|
| 30 |
def _write_adapter(path: Path) -> None: |
| 31 |
path.mkdir(parents=True) |
| 32 |
(path / "adapter_config.json").write_text("{}", encoding="utf-8") |
| 33 |
(path / "adapter_model.safetensors").write_bytes(b"adapter") |
| 34 |
|
| 35 |
|
| 36 |
def _fake_stage_mlx(src: Path, dst: Path, *, base_hf_id: str) -> Path: |
| 37 |
assert src.exists() |
| 38 |
assert base_hf_id == _SPEC.hf_id |
| 39 |
dst.mkdir(parents=True, exist_ok=True) |
| 40 |
(dst / "adapter_config.json").write_text("{}", encoding="utf-8") |
| 41 |
(dst / "adapters.safetensors").write_bytes(b"mlx-adapter") |
| 42 |
return dst |
| 43 |
|
| 44 |
|
| 45 |
def _setup_flat_store(tmp_path: Path) -> object: |
| 46 |
store = for_dlm("01MLXTEST", home=tmp_path) |
| 47 |
store.ensure_layout() |
| 48 |
save_manifest(store.manifest, Manifest(dlm_id="01MLXTEST", base_model=_SPEC.key)) |
| 49 |
adapter = store.adapter_version(3) |
| 50 |
_write_adapter(adapter) |
| 51 |
store.set_current_adapter(adapter) |
| 52 |
return store |
| 53 |
|
| 54 |
|
| 55 |
def _setup_named_store(tmp_path: Path) -> object: |
| 56 |
store = for_dlm("01MLXMULTI", home=tmp_path) |
| 57 |
store.ensure_layout() |
| 58 |
save_manifest(store.manifest, Manifest(dlm_id="01MLXMULTI", base_model=_SPEC.key)) |
| 59 |
knowledge = store.adapter_version_for("knowledge", 2) |
| 60 |
tone = store.adapter_version_for("tone", 4) |
| 61 |
_write_adapter(knowledge) |
| 62 |
_write_adapter(tone) |
| 63 |
store.set_current_adapter_for("knowledge", knowledge) |
| 64 |
store.set_current_adapter_for("tone", tone) |
| 65 |
return store |
| 66 |
|
| 67 |
|
| 68 |
class TestPrepareMlxServeExport: |
| 69 |
def test_prepare_method_is_not_used_directly(self) -> None: |
| 70 |
with pytest.raises(NotImplementedError, match="prepare_mlx_serve_export"): |
| 71 |
MLX_SERVE_TARGET.prepare(object()) |
| 72 |
|
| 73 |
def test_prepare_writes_launch_script_and_manifest( |
| 74 |
self, tmp_path: Path, monkeypatch: object |
| 75 |
) -> None: |
| 76 |
store = _setup_flat_store(tmp_path) |
| 77 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 78 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 79 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.stage_mlx_adapter_dir", _fake_stage_mlx) |
| 80 |
|
| 81 |
prepared = prepare_mlx_serve_export( |
| 82 |
store=store, |
| 83 |
spec=_SPEC, |
| 84 |
adapter_name=None, |
| 85 |
adapter_path_override=None, |
| 86 |
declared_adapter_names=None, |
| 87 |
) |
| 88 |
manifest_path = finalize_mlx_serve_export( |
| 89 |
store=store, |
| 90 |
spec=_SPEC, |
| 91 |
prepared=prepared, |
| 92 |
smoke_output_first_line="hello from mlx", |
| 93 |
adapter_name=None, |
| 94 |
adapter_mix=None, |
| 95 |
) |
| 96 |
|
| 97 |
assert prepared.launch_script_path is not None |
| 98 |
assert prepared.launch_script_path.name == LAUNCH_SCRIPT_FILENAME |
| 99 |
script = prepared.launch_script_path.read_text(encoding="utf-8") |
| 100 |
assert script.startswith("#!/usr/bin/env bash\nset -euo pipefail\n") |
| 101 |
assert "python -m mlx_lm.server" in script |
| 102 |
assert f"--model {_SPEC.hf_id}" in script |
| 103 |
assert '--adapter-path "$SCRIPT_DIR/adapter"' in script |
| 104 |
|
| 105 |
export_manifest = load_export_manifest(prepared.export_dir) |
| 106 |
assert manifest_path == prepared.manifest_path |
| 107 |
assert export_manifest.target == "mlx-serve" |
| 108 |
assert export_manifest.quant == "hf" |
| 109 |
assert export_manifest.adapter_version == 3 |
| 110 |
assert any(artifact.path == "mlx_serve_launch.sh" for artifact in export_manifest.artifacts) |
| 111 |
assert any( |
| 112 |
artifact.path == "adapter/adapters.safetensors" |
| 113 |
for artifact in export_manifest.artifacts |
| 114 |
) |
| 115 |
|
| 116 |
store_manifest = load_manifest(store.manifest) |
| 117 |
assert store_manifest.exports[-1].target == "mlx-serve" |
| 118 |
assert store_manifest.exports[-1].quant == "hf" |
| 119 |
assert store_manifest.exports[-1].smoke_output_first_line == "hello from mlx" |
| 120 |
|
| 121 |
def test_prepare_replaces_stale_staged_adapter_dir( |
| 122 |
self, tmp_path: Path, monkeypatch: object |
| 123 |
) -> None: |
| 124 |
store = _setup_flat_store(tmp_path) |
| 125 |
export_dir = store.exports / "mlx-serve" |
| 126 |
stale_dir = export_dir / "adapter" |
| 127 |
stale_dir.mkdir(parents=True) |
| 128 |
(stale_dir / "stale.txt").write_text("stale", encoding="utf-8") |
| 129 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 130 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 131 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.stage_mlx_adapter_dir", _fake_stage_mlx) |
| 132 |
|
| 133 |
prepared = prepare_mlx_serve_export( |
| 134 |
store=store, |
| 135 |
spec=_SPEC, |
| 136 |
adapter_name=None, |
| 137 |
adapter_path_override=None, |
| 138 |
declared_adapter_names=None, |
| 139 |
) |
| 140 |
|
| 141 |
assert prepared.launch_script_path is not None |
| 142 |
assert not (prepared.export_dir / "adapter" / "stale.txt").exists() |
| 143 |
assert (prepared.export_dir / "adapter" / "adapters.safetensors").exists() |
| 144 |
|
| 145 |
def test_multi_adapter_export_requires_explicit_selection( |
| 146 |
self, tmp_path: Path, monkeypatch: object |
| 147 |
) -> None: |
| 148 |
store = _setup_named_store(tmp_path) |
| 149 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 150 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 151 |
|
| 152 |
with pytest.raises(ExportError, match="one adapter at a time"): |
| 153 |
prepare_mlx_serve_export( |
| 154 |
store=store, |
| 155 |
spec=_SPEC, |
| 156 |
adapter_name=None, |
| 157 |
adapter_path_override=None, |
| 158 |
declared_adapter_names=("knowledge", "tone"), |
| 159 |
) |
| 160 |
|
| 161 |
def test_refuses_without_apple_silicon_runtime( |
| 162 |
self, tmp_path: Path, monkeypatch: object |
| 163 |
) -> None: |
| 164 |
store = _setup_flat_store(tmp_path) |
| 165 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: False) |
| 166 |
|
| 167 |
with pytest.raises(ExportError, match="Apple Silicon"): |
| 168 |
prepare_mlx_serve_export( |
| 169 |
store=store, |
| 170 |
spec=_SPEC, |
| 171 |
adapter_name=None, |
| 172 |
adapter_path_override=None, |
| 173 |
declared_adapter_names=None, |
| 174 |
) |
| 175 |
|
| 176 |
def test_refuses_without_mlx_extra(self, tmp_path: Path, monkeypatch: object) -> None: |
| 177 |
store = _setup_flat_store(tmp_path) |
| 178 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 179 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: False) |
| 180 |
|
| 181 |
with pytest.raises(ExportError, match="mlx extra"): |
| 182 |
prepare_mlx_serve_export( |
| 183 |
store=store, |
| 184 |
spec=_SPEC, |
| 185 |
adapter_name=None, |
| 186 |
adapter_path_override=None, |
| 187 |
declared_adapter_names=None, |
| 188 |
) |
| 189 |
|
| 190 |
def test_missing_named_adapter_raises(self, tmp_path: Path, monkeypatch: object) -> None: |
| 191 |
store = _setup_named_store(tmp_path) |
| 192 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 193 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 194 |
|
| 195 |
with pytest.raises(ExportError, match="no current adapter under"): |
| 196 |
prepare_mlx_serve_export( |
| 197 |
store=store, |
| 198 |
spec=_SPEC, |
| 199 |
adapter_name="missing", |
| 200 |
adapter_path_override=None, |
| 201 |
declared_adapter_names=None, |
| 202 |
) |
| 203 |
|
| 204 |
def test_named_adapter_export_uses_named_dir(self, tmp_path: Path, monkeypatch: object) -> None: |
| 205 |
store = _setup_named_store(tmp_path) |
| 206 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 207 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 208 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.stage_mlx_adapter_dir", _fake_stage_mlx) |
| 209 |
|
| 210 |
prepared = prepare_mlx_serve_export( |
| 211 |
store=store, |
| 212 |
spec=_SPEC, |
| 213 |
adapter_name="knowledge", |
| 214 |
adapter_path_override=None, |
| 215 |
declared_adapter_names=None, |
| 216 |
) |
| 217 |
|
| 218 |
assert str(prepared.extras["adapter_dir"]).endswith("knowledge") |
| 219 |
assert prepared.extras["adapter_version"] == 2 |
| 220 |
|
| 221 |
def test_missing_adapter_override_raises(self, tmp_path: Path, monkeypatch: object) -> None: |
| 222 |
store = _setup_flat_store(tmp_path) |
| 223 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 224 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 225 |
|
| 226 |
with pytest.raises(ExportError, match="adapter_path_override .* does not exist"): |
| 227 |
prepare_mlx_serve_export( |
| 228 |
store=store, |
| 229 |
spec=_SPEC, |
| 230 |
adapter_name=None, |
| 231 |
adapter_path_override=tmp_path / "missing", |
| 232 |
declared_adapter_names=None, |
| 233 |
) |
| 234 |
|
| 235 |
def test_existing_adapter_override_uses_mixed_dir( |
| 236 |
self, tmp_path: Path, monkeypatch: object |
| 237 |
) -> None: |
| 238 |
store = _setup_flat_store(tmp_path) |
| 239 |
override = tmp_path / "custom-adapter" |
| 240 |
_write_adapter(override) |
| 241 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 242 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 243 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.stage_mlx_adapter_dir", _fake_stage_mlx) |
| 244 |
|
| 245 |
prepared = prepare_mlx_serve_export( |
| 246 |
store=store, |
| 247 |
spec=_SPEC, |
| 248 |
adapter_name=None, |
| 249 |
adapter_path_override=override, |
| 250 |
declared_adapter_names=None, |
| 251 |
) |
| 252 |
|
| 253 |
assert str(prepared.extras["adapter_dir"]).endswith("mixed") |
| 254 |
assert prepared.extras["adapter_version"] == 1 |
| 255 |
|
| 256 |
def test_missing_default_adapter_raises(self, tmp_path: Path, monkeypatch: object) -> None: |
| 257 |
store = for_dlm("01EMPTYMLX", home=tmp_path) |
| 258 |
store.ensure_layout() |
| 259 |
save_manifest(store.manifest, Manifest(dlm_id="01EMPTYMLX", base_model=_SPEC.key)) |
| 260 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 261 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 262 |
|
| 263 |
with pytest.raises(ExportError, match="no current adapter under"): |
| 264 |
prepare_mlx_serve_export( |
| 265 |
store=store, |
| 266 |
spec=_SPEC, |
| 267 |
adapter_name=None, |
| 268 |
adapter_path_override=None, |
| 269 |
declared_adapter_names=None, |
| 270 |
) |
| 271 |
|
| 272 |
|
| 273 |
class TestMlxServeSmoke: |
| 274 |
def test_smoke_uses_absolute_runtime_paths(self, tmp_path: Path, monkeypatch: object) -> None: |
| 275 |
store = _setup_flat_store(tmp_path) |
| 276 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 277 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 278 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.stage_mlx_adapter_dir", _fake_stage_mlx) |
| 279 |
prepared = prepare_mlx_serve_export( |
| 280 |
store=store, |
| 281 |
spec=_SPEC, |
| 282 |
adapter_name=None, |
| 283 |
adapter_path_override=None, |
| 284 |
declared_adapter_names=None, |
| 285 |
) |
| 286 |
seen: list[list[str]] = [] |
| 287 |
|
| 288 |
def _fake_smoke(argv: list[str], **_: object) -> str: |
| 289 |
seen.append(list(argv)) |
| 290 |
return "mlx replied" |
| 291 |
|
| 292 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.smoke_openai_compat_server", _fake_smoke) |
| 293 |
|
| 294 |
result = MLX_SERVE_TARGET.smoke_test(prepared) |
| 295 |
|
| 296 |
assert result.attempted is True |
| 297 |
assert result.ok is True |
| 298 |
assert result.detail == "mlx replied" |
| 299 |
argv = seen[0] |
| 300 |
assert argv[:3] == ["python", "-m", "mlx_lm.server"] |
| 301 |
assert "$SCRIPT_DIR" not in " ".join(argv) |
| 302 |
assert _SPEC.hf_id in argv |
| 303 |
assert str(prepared.export_dir / "adapter") in argv |
| 304 |
|
| 305 |
def test_smoke_failure_returns_failed_result(self, tmp_path: Path, monkeypatch: object) -> None: |
| 306 |
store = _setup_flat_store(tmp_path) |
| 307 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.is_apple_silicon", lambda: True) |
| 308 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.mlx_available", lambda: True) |
| 309 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.stage_mlx_adapter_dir", _fake_stage_mlx) |
| 310 |
prepared = prepare_mlx_serve_export( |
| 311 |
store=store, |
| 312 |
spec=_SPEC, |
| 313 |
adapter_name=None, |
| 314 |
adapter_path_override=None, |
| 315 |
declared_adapter_names=None, |
| 316 |
) |
| 317 |
|
| 318 |
def _fake_smoke(argv: list[str], **_: object) -> str: |
| 319 |
_ = argv |
| 320 |
raise TargetSmokeError("boom") |
| 321 |
|
| 322 |
monkeypatch.setattr("dlm.export.targets.mlx_serve.smoke_openai_compat_server", _fake_smoke) |
| 323 |
|
| 324 |
result = MLX_SERVE_TARGET.smoke_test(prepared) |
| 325 |
|
| 326 |
assert result.attempted is True |
| 327 |
assert result.ok is False |
| 328 |
assert result.detail == "boom" |
| 329 |
|
| 330 |
|
| 331 |
class TestMlxServeHelpers: |
| 332 |
def test_quote_script_arg_and_int_validation(self) -> None: |
| 333 |
assert _quote_script_arg("$SCRIPT_DIR/adapter") == '"$SCRIPT_DIR/adapter"' |
| 334 |
assert _quote_script_arg("plain value") == "'plain value'" |
| 335 |
|
| 336 |
prepared = TargetResult( |
| 337 |
name="mlx-serve", |
| 338 |
export_dir=Path("/tmp/export"), |
| 339 |
manifest_path=Path("/tmp/export/export_manifest.json"), |
| 340 |
extras={"adapter_version": "bad"}, |
| 341 |
) |
| 342 |
with pytest.raises(ExportError, match="missing int extra"): |
| 343 |
_require_prepared_int(prepared, "adapter_version") |
| 344 |
|
| 345 |
def test_string_and_path_validation(self) -> None: |
| 346 |
prepared = TargetResult( |
| 347 |
name="mlx-serve", |
| 348 |
export_dir=Path("/tmp/export"), |
| 349 |
manifest_path=Path("/tmp/export/export_manifest.json"), |
| 350 |
extras={"model": "", "adapter_dir": "bad"}, |
| 351 |
) |
| 352 |
|
| 353 |
with pytest.raises(ExportError, match="missing string extra"): |
| 354 |
_require_prepared_str(prepared, "model") |
| 355 |
with pytest.raises(ExportError, match="missing Path extra"): |
| 356 |
_require_prepared_path(prepared, "adapter_dir") |
| 357 |
|
| 358 |
def test_version_from_dir_name_defaults_for_non_version_dirs(self) -> None: |
| 359 |
assert _version_from_dir_name(Path("custom-adapter")) == 1 |