diff --git a/libs/infinity_emb/infinity_emb/engine.py b/libs/infinity_emb/infinity_emb/engine.py
index 0db14ea5..d23a1bf4 100644
--- a/libs/infinity_emb/infinity_emb/engine.py
+++ b/libs/infinity_emb/infinity_emb/engine.py
@@ -336,7 +336,7 @@ async def classify(
         return await self[model].classify(sentences=sentences, raw_scores=raw_scores)
 
     async def image_embed(
-        self, model: str, images: list[str]
+        self, *, model: str, images: list[str]
     ) -> tuple[list[EmbeddingReturnType], int]:
         """embed multiple images
 
diff --git a/libs/infinity_emb/infinity_emb/sync_engine.py b/libs/infinity_emb/infinity_emb/sync_engine.py
index 7a9774ab..cf49405d 100644
--- a/libs/infinity_emb/infinity_emb/sync_engine.py
+++ b/libs/infinity_emb/infinity_emb/sync_engine.py
@@ -99,14 +99,14 @@ def async_run(
 
 @add_start_docstrings(AsyncEngineArray.__doc__)
 class SyncEngineArray(AsyncLifeMixin):
-    def __init__(self, engine_args: list[EngineArgs]):
+    def __init__(self, _engine_args_array: list[EngineArgs]):
         super().__init__()
-        self.async_engine_array = AsyncEngineArray.from_args(engine_args)
+        self.async_engine_array = AsyncEngineArray.from_args(_engine_args_array)
         self.async_run(self.async_engine_array.astart).result()
 
     @classmethod
-    def from_args(cls, engine_args: list[EngineArgs]) -> "SyncEngineArray":
-        return cls(engine_args)
+    def from_args(cls, engine_args_array: list[EngineArgs]) -> "SyncEngineArray":
+        return cls(_engine_args_array=engine_args_array)
 
     @property
     def is_running(self):
@@ -128,17 +128,26 @@ def embed(self, *, model: str, sentences: list[str]):
         )
 
     @add_start_docstrings(AsyncEngineArray.rerank.__doc__)
-    def rerank(self, *, model: str, query: str, docs: list[str]):
+    def rerank(
+        self, *, model: str, query: str, docs: list[str], raw_scores: bool = False
+    ):
         """sync interface of AsyncEngineArray"""
         return self.async_run(
-            self.async_engine_array.rerank, model=model, query=query, docs=docs
+            self.async_engine_array.rerank,
+            model=model,
+            query=query,
+            docs=docs,
+            raw_scores=raw_scores,
         )
 
     @add_start_docstrings(AsyncEngineArray.classify.__doc__)
-    def classify(self, *, model: str, sentences: str):
+    def classify(self, *, model: str, sentences: list[str], raw_scores: bool = False):
         """sync interface of AsyncEngineArray"""
         return self.async_run(
-            self.async_engine_array.classify, model=model, sentences=sentences
+            self.async_engine_array.classify,
+            model=model,
+            sentences=sentences,
+            raw_scores=raw_scores,
         )
 
     @add_start_docstrings(AsyncEngineArray.image_embed.__doc__)
diff --git a/libs/infinity_emb/pyproject.toml b/libs/infinity_emb/pyproject.toml
index a6778e79..ccccc6e8 100644
--- a/libs/infinity_emb/pyproject.toml
+++ b/libs/infinity_emb/pyproject.toml
@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "infinity_emb"
-version = "0.0.48"
+version = "0.0.49"
 description = "Infinity is a high-throughput, low-latency REST API for serving vector embeddings, supporting a wide range of sentence-transformer models and frameworks."
 authors = ["michaelfeil <me@michaelfeil.eu>"]
 license = "MIT"
diff --git a/libs/infinity_emb/tests/conftest.py b/libs/infinity_emb/tests/conftest.py
index f46fecc6..4fec5bd5 100644
--- a/libs/infinity_emb/tests/conftest.py
+++ b/libs/infinity_emb/tests/conftest.py
@@ -9,6 +9,8 @@
 pytest.DEFAULT_RERANKER_MODEL = "mixedbread-ai/mxbai-rerank-xsmall-v1"
 pytest.DEFAULT_CLASSIFIER_MODEL = "SamLowe/roberta-base-go_emotions"
 
+pytest.ENGINE_METHODS = ["embed", "image_embed", "classify", "rerank"]
+
 
 @pytest.fixture
 def anyio_backend():
diff --git a/libs/infinity_emb/tests/unit_test/test_engine.py b/libs/infinity_emb/tests/unit_test/test_engine.py
index d833f443..89b1a3df 100644
--- a/libs/infinity_emb/tests/unit_test/test_engine.py
+++ b/libs/infinity_emb/tests/unit_test/test_engine.py
@@ -1,4 +1,5 @@
 import asyncio
+import inspect
 import sys
 
 import numpy as np
@@ -6,7 +7,7 @@
 import torch
 from sentence_transformers import CrossEncoder  # type: ignore[import-untyped]
 
-from infinity_emb import AsyncEmbeddingEngine, EngineArgs
+from infinity_emb import AsyncEmbeddingEngine, AsyncEngineArray, EngineArgs
 from infinity_emb.primitives import (
     Device,
     EmbeddingDtype,
@@ -251,3 +252,14 @@ async def test_async_api_failing_revision():
                 revision="a32952c6d05d45f64f9f709a092c00839bcfe70a",
             )
         )
+
+
+@pytest.mark.parametrize("method_name", list(pytest.ENGINE_METHODS))  # type: ignore
+def test_args_between_array_and_engine_same(method_name: str):
+    array_method = inspect.getfullargspec(getattr(AsyncEngineArray, method_name))
+    engine_method = inspect.getfullargspec(getattr(AsyncEmbeddingEngine, method_name))
+
+    assert "model" in array_method.kwonlyargs
+    assert sorted(array_method.args + array_method.kwonlyargs) == sorted(
+        engine_method.args + engine_method.kwonlyargs + ["model"]
+    )
diff --git a/libs/infinity_emb/tests/unit_test/test_sync_engine.py b/libs/infinity_emb/tests/unit_test/test_sync_engine.py
index 6b8f1053..6e238012 100644
--- a/libs/infinity_emb/tests/unit_test/test_sync_engine.py
+++ b/libs/infinity_emb/tests/unit_test/test_sync_engine.py
@@ -1,8 +1,9 @@
+import inspect
 from uuid import uuid4
 
 import pytest
 
-from infinity_emb import EngineArgs, SyncEngineArray
+from infinity_emb import AsyncEngineArray, EngineArgs, SyncEngineArray
 
 
 def test_sync_engine():
@@ -75,5 +76,16 @@ def test_sync_engine_on_model(model_id, method: str, payload: dict):
         s_eng_array.stop()
 
 
+@pytest.mark.parametrize("method_name", list(pytest.ENGINE_METHODS) + ["from_args"])  # type: ignore
+def test_args_between_sync_and_async_same(method_name: str):
+    sync_method = inspect.getfullargspec(getattr(SyncEngineArray, method_name))
+    async_method = inspect.getfullargspec(getattr(AsyncEngineArray, method_name))
+    if method_name in list(pytest.ENGINE_METHODS):  # type: ignore
+        assert "model" in sync_method.kwonlyargs
+        assert "model" in async_method.kwonlyargs
+    assert sync_method.args == async_method.args
+    assert sync_method.kwonlyargs == async_method.kwonlyargs
+
+
 if __name__ == "__main__":
     test_sync_engine()
diff --git a/libs/simpleinference/poetry.lock b/libs/simpleinference/poetry.lock
index 8212f7e7..4be8c0e4 100644
--- a/libs/simpleinference/poetry.lock
+++ b/libs/simpleinference/poetry.lock
@@ -933,22 +933,20 @@ files = [
 
 [[package]]
 name = "infinity-emb"
-version = "0.0.47"
+version = "0.0.49"
 description = "Infinity is a high-throughput, low-latency REST API for serving vector embeddings, supporting a wide range of sentence-transformer models and frameworks."
 optional = false
-python-versions = "<4,>=3.9"
-files = [
-    {file = "infinity_emb-0.0.47-py3-none-any.whl", hash = "sha256:53790664be0c677d72009869d73cf9e03cc107694f2c128ed782ef24138b8ed2"},
-    {file = "infinity_emb-0.0.47.tar.gz", hash = "sha256:e9835f37301a16f73ea282cf511917d9a65547f9acb408a8d4245ca7de9d7cb7"},
-]
+python-versions = ">=3.9,<4"
+files = []
+develop = false
 
 [package.dependencies]
 hf_transfer = ">=0.1.5"
 huggingface_hub = "*"
 numpy = ">=1.20.0,<2"
-optimum = {version = ">=1.16.2", extras = ["onnxruntime"], optional = true, markers = "extra == \"optimum\" or extra == \"all\""}
-pillow = {version = "*", optional = true, markers = "extra == \"vision\" or extra == \"all\""}
-timm = {version = "*", optional = true, markers = "extra == \"vision\" or extra == \"all\""}
+optimum = {version = ">=1.16.2", extras = ["onnxruntime"], optional = true}
+pillow = {version = "*", optional = true}
+timm = {version = "*", optional = true}
 
 [package.extras]
 all = ["ctranslate2 (>=4.0.0,<5.0.0)", "diskcache", "einops", "fastapi (>=0.103.2)", "optimum[onnxruntime] (>=1.16.2)", "orjson (>=3.9.8,!=3.10.0)", "pillow", "prometheus-fastapi-instrumentator (>=6.1.0)", "pydantic (>=2.4.0,<3)", "rich (>=13,<14)", "sentence-transformers (>=3.0.1,<4.0.0)", "timm", "torch (>=2.2.1)", "typer[all] (>=0.9.0,<0.10.0)", "uvicorn[standard] (>=0.23.2,<0.24.0)"]
@@ -963,6 +961,10 @@ tensorrt = ["tensorrt (>=8.6.1,<9.0.0)"]
 torch = ["sentence-transformers (>=3.0.1,<4.0.0)", "torch (>=2.2.1)"]
 vision = ["pillow", "timm"]
 
+[package.source]
+type = "directory"
+url = "../infinity_emb"
+
 [[package]]
 name = "iniconfig"
 version = "2.0.0"
@@ -3125,4 +3127,4 @@ multidict = ">=4.0"
 [metadata]
 lock-version = "2.0"
 python-versions = ">=3.9,<4"
-content-hash = "9f8306b6bb33ee205dafb73816204996153c371d643786bbd1f7e516b34aa988"
+content-hash = "090cebaf1aba5d7cfb553d30fe32f524cc2ca990d9352ea066deca74fe510884"
diff --git a/libs/simpleinference/pyproject.toml b/libs/simpleinference/pyproject.toml
index 644e9d4b..1ab7fc1e 100644
--- a/libs/simpleinference/pyproject.toml
+++ b/libs/simpleinference/pyproject.toml
@@ -12,7 +12,7 @@ priority = "explicit"
 
 [tool.poetry.dependencies]
 python = ">=3.9,<4"
-infinity_emb = {version = "0.0.47", extras = ["optimum","vision"]}
+infinity_emb = {path = "../infinity_emb", extras = ["optimum","vision"]}
 
 [tool.poetry.group.test.dependencies]
 pytest = "^7.0.0"
diff --git a/libs/simpleinference/simpleinference/infer.py b/libs/simpleinference/simpleinference/infer.py
index dcb7d172..051d58ed 100644
--- a/libs/simpleinference/simpleinference/infer.py
+++ b/libs/simpleinference/simpleinference/infer.py
@@ -48,7 +48,9 @@ def __init__(
             )
             for m, e, d, edt in zip(model_id, engine, device, embedding_dtype)
         ]
-        self._engine_array = SyncEngineArray.from_args(engine_args=self._engine_args)
+        self._engine_array = SyncEngineArray.from_args(
+            engine_args_array=self._engine_args
+        )
 
     def stop(self):
         self._engine_array.stop()