fix(tests): building not being isolated

We will need to fix this from BentoML

Signed-off-by: aarnphm-ec2-dev <29749331+aarnphm@users.noreply.github.com>

pyproject.toml

@@ -72,12 +72,12 @@ all = [
     "openllm[falcon]",
     "openllm[mpt]",
     "openllm[starcoder]",
-    "openllm[openai]",
     "openllm[flan-t5]",
     "openllm[fine-tune]",
     "openllm[agents]",
     "openllm[playground]",
     "openllm[ggml]",
+    "openllm[openai]",
 ]
 chatglm = ["cpm-kernels", "sentencepiece"]
 falcon = ["einops", "xformers", "safetensors"]

src/openllm/_configuration.py

@@ -1622,6 +1622,8 @@ class LLMConfig(_ConfigAttr):
             **attrs: The attributes to be added to the new class. This will override
                      any existing attributes with the same name.
         """
+        if not hasattr(cls, "__config__"):
+            raise ValueError("Cannot derivate a LLMConfig without __config__")
         _new_cfg = {k: v for k, v in attrs.items() if k in attr.fields_dict(_ModelSettingsAttr)}
         attrs = {k: v for k, v in attrs.items() if k not in _new_cfg}
         new_cls = types.new_class(

src/openllm/cli.py

@@ -1403,50 +1403,6 @@ def _start(
     )
 
 
-@overload
-def _build(
-    model_name: str,
-    /,
-    *,
-    model_id: str | None = ...,
-    model_version: str | None = ...,
-    quantize: t.Literal["int8", "int4", "gptq"] | None = ...,
-    bettertransformer: bool | None = ...,
-    adapter_map: dict[str, str | None] | None = ...,
-    build_ctx: str | None = ...,
-    enable_features: tuple[str, ...] | None = ...,
-    workers_per_resource: int | float | None = ...,
-    runtime: t.Literal["ggml", "transformers"] = ...,
-    dockerfile_template: str | None = ...,
-    overwrite: bool = ...,
-    format: t.Literal["bento"] = "bento",
-    additional_args: list[str] | None = ...,
-) -> bentoml.Bento:
-    ...
-
-
-@overload
-def _build(
-    model_name: str,
-    /,
-    *,
-    model_id: str | None = ...,
-    model_version: str | None = ...,
-    quantize: t.Literal["int8", "int4", "gptq"] | None = ...,
-    bettertransformer: bool | None = ...,
-    adapter_map: dict[str, str | None] | None = ...,
-    build_ctx: str | None = ...,
-    enable_features: tuple[str, ...] | None = ...,
-    workers_per_resource: int | float | None = ...,
-    runtime: t.Literal["ggml", "transformers"] = ...,
-    dockerfile_template: str | None = ...,
-    overwrite: bool = ...,
-    format: t.Literal["container"] = ...,
-    additional_args: list[str] | None = ...,
-) -> str:
-    ...
-
-
 def _build(
     model_name: str,
     /,
@@ -1464,7 +1420,7 @@ def _build(
     overwrite: bool = False,
     format: t.Literal["bento", "container"] = "bento",
     additional_args: list[str] | None = None,
-) -> bentoml.Bento | str:
+) -> bentoml.Bento:
     """Package a LLM into a Bento.
 
     The LLM will be built into a BentoService with the following structure:
@@ -1689,8 +1645,6 @@ def build_command(
     > NOTE: To run a container built from this Bento with GPU support, make sure
     > to have https://github.com/NVIDIA/nvidia-container-toolkit install locally.
     """
-    from bentoml_cli.cli import cli as bentoml_cli
-
     from ._package import create_bento
 
     adapter_map: dict[str, str | None] | None = None
@@ -1834,16 +1788,15 @@ def build_command(
     else:
         _echo(bento.tag)
 
-    if format == "bento":
-        return bento
+    if format == "container":
+        backend = os.getenv("BENTOML_CONTAINERIZE_BACKEND", "docker")
+        _echo(f"Building {bento} into a LLMContainer using backend '{backend}'", fg="magenta")
+        if not bentoml.container.health(backend):
+            raise OpenLLMException(f"Failed to use backend {backend}")
 
-    backend = os.getenv("BENTOML_CONTAINERIZE_BACKEND", "docker")
-    _echo(f"\nBuilding {bento} into a LLMContainer using backend '{backend}'", fg="magenta")
-    args = [str(bento.tag), "--backend", backend]
-    if get_debug_mode():
-        args.extend(["--opt", "progress=plain"])
-    bentoml_cli.commands["containerize"].main(standalone_mode=False, args=args)
-    return str(bento.tag)
+        bentoml.container.build(str(bento.tag), backend=backend, features=("grpc",))
+
+    return bento
 
 
 @overload

tests/package_test.py

@@ -42,8 +42,6 @@ def test_general_build_with_internal_testing():
 
 
 def test_general_build_from_local(tmp_path_factory: pytest.TempPathFactory):
-    bento_store = BentoMLContainer.bento_store.get()
-
     local_path = tmp_path_factory.mktemp("local_t5")
     llm = openllm.AutoLLM.for_model("flan-t5", model_id=HF_INTERNAL_T5_TESTING, ensure_available=True)
 
@@ -52,11 +50,7 @@ def test_general_build_from_local(tmp_path_factory: pytest.TempPathFactory):
 
     llm.save_pretrained(local_path)
 
-    bento = openllm.build("flan-t5", model_id=local_path.resolve().__fspath__(), model_version="1")
-    assert len(bento_store.list(bento.tag)) == 1
-
-    bento = openllm.build("flan-t5", model_id=local_path.resolve().__fspath__(), model_version="1")
-    assert len(bento_store.list(bento.tag)) == 1
+    assert openllm.build("flan-t5", model_id=local_path.resolve().__fspath__(), model_version="local")
 
 
 @pytest.fixture(name="dockerfile_template")
@@ -70,9 +64,4 @@ def fixture_dockerfile_template(tmp_path_factory: pytest.TempPathFactory):
 
 @pytest.mark.usefixtures("dockerfile_template")
 def test_build_with_custom_dockerfile(dockerfile_template: Path):
-    assert openllm.build(
-        "flan-t5",
-        model_id=HF_INTERNAL_T5_TESTING,
-        overwrite=True,
-        dockerfile_template=str(dockerfile_template),
-    )
+    assert openllm.build("flan-t5", model_id=HF_INTERNAL_T5_TESTING, dockerfile_template=str(dockerfile_template))