refactor: modularity P1-P3

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: leebeanbin

src/beanllm/__init__.py

@@ -19,18 +19,18 @@
     pip install transformers torch torchaudio nemo_toolkit
 """
 
-import logging
 import time
 from pathlib import Path
 from typing import Dict, Union
 
 import numpy as np
 
 from beanllm.domain.audio.models import STTConfig
+from beanllm.utils.logging import get_logger
 
 from .base import BaseSTTEngine
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
 
 # NeMo toolkit 설치 여부 체크
 try:

src/beanllm/_lazy_imports.py

@@ -15,18 +15,18 @@
     pip install transformers torch torchaudio
 """
 
-import logging
 import time
 from pathlib import Path
 from typing import Dict, Union
 
 import numpy as np
 
 from beanllm.domain.audio.models import STTConfig
+from beanllm.utils.logging import get_logger
 
 from .base import BaseSTTEngine
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
 
 # transformers 설치 여부 체크
 try:

src/beanllm/domain/audio/engines/canary_engine.py

@@ -23,18 +23,18 @@
     pip install transformers torch torchaudio
 """
 
-import logging
 import time
 from pathlib import Path
 from typing import Dict, Union
 
 import numpy as np
 
 from beanllm.domain.audio.models import STTConfig
+from beanllm.utils.logging import get_logger
 
 from .base import BaseSTTEngine
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
 
 # transformers 설치 여부 체크
 try:

src/beanllm/domain/audio/engines/distil_whisper_engine.py

@@ -21,18 +21,18 @@
     pip install transformers torch torchaudio
 """
 
-import logging
 import time
 from pathlib import Path
 from typing import Dict, Union
 
 import numpy as np
 
 from beanllm.domain.audio.models import STTConfig
+from beanllm.utils.logging import get_logger
 
 from .base import BaseSTTEngine
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
 
 # transformers 설치 여부 체크
 try:

src/beanllm/domain/audio/engines/granite_engine.py

@@ -15,18 +15,18 @@
     pip install transformers torch torchaudio nemo_toolkit
 """
 
-import logging
 import time
 from pathlib import Path
 from typing import Dict, Union
 
 import numpy as np
 
 from beanllm.domain.audio.models import STTConfig
+from beanllm.utils.logging import get_logger
 
 from .base import BaseSTTEngine
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
 
 # NeMo toolkit 설치 여부 체크
 try:

src/beanllm/domain/audio/engines/moonshine_engine.py

@@ -30,18 +30,18 @@
     pip install funasr modelscope torch torchaudio
 """
 
-import logging
 import time
 from pathlib import Path
 from typing import Dict, Union
 
 import numpy as np
 
 from beanllm.domain.audio.models import STTConfig
+from beanllm.utils.logging import get_logger
 
 from .base import BaseSTTEngine
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
 
 # FunASR 설치 여부 체크
 try:

src/beanllm/domain/audio/engines/parakeet_engine.py

@@ -15,18 +15,18 @@
     pip install transformers torch torchaudio
 """
 
-import logging
 import time
 from pathlib import Path
 from typing import Dict, Union
 
 import numpy as np
 
 from beanllm.domain.audio.models import STTConfig
+from beanllm.utils.logging import get_logger
 
 from .base import BaseSTTEngine
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
 
 # transformers 설치 여부 체크
 try:

src/beanllm/domain/audio/engines/sensevoice_engine.py

@@ -0,0 +1,95 @@
+"""
+DeepEval Metrics - 메트릭 생성 및 설정
+
+DeepEval 메트릭의 팩토리 로직과 설정 상수를 제공합니다.
+deepeval_wrapper.py에서 평가 실행 시 이 모듈을 사용합니다.
+"""
+
+from typing import Any, Dict, Type
+
+# DeepEval 메트릭 이름 → 설명 (list_tasks 등에서 사용)
+DEEPEVAL_METRIC_DESCRIPTIONS: Dict[str, str] = {
+    "answer_relevancy": "답변이 질문과 얼마나 관련있는지",
+    "faithfulness": "답변이 컨텍스트에 충실한지 (Hallucination 방지)",
+    "contextual_precision": "검색된 컨텍스트의 정밀도",
+    "contextual_recall": "검색된 컨텍스트의 재현율",
+    "hallucination": "환각 감지",
+    "toxicity": "독성 평가",
+    "bias": "편향 평가",
+    "summarization": "요약 품질",
+    "geval": "커스텀 평가 기준",
+}
+
+
+def get_deepeval_metric_map() -> Dict[str, Type[Any]]:
+    """
+    DeepEval 메트릭 클래스 매핑 반환.
+
+    Returns:
+        metric_name → Metric class 매핑
+    """
+    from deepeval.metrics import (  # type: ignore[import-untyped]
+        AnswerRelevancyMetric,
+        BiasMetric,
+        ContextualPrecisionMetric,
+        ContextualRecallMetric,
+        FaithfulnessMetric,
+        GEval,
+        HallucinationMetric,
+        SummarizationMetric,
+        ToxicityMetric,
+    )
+
+    return {
+        "answer_relevancy": AnswerRelevancyMetric,
+        "faithfulness": FaithfulnessMetric,
+        "contextual_precision": ContextualPrecisionMetric,
+        "contextual_recall": ContextualRecallMetric,
+        "hallucination": HallucinationMetric,
+        "toxicity": ToxicityMetric,
+        "bias": BiasMetric,
+        "summarization": SummarizationMetric,
+        "geval": GEval,
+    }
+
+
+def create_deepeval_metric(
+    metric_name: str,
+    model: str = "gpt-4o-mini",
+    threshold: float = 0.5,
+    include_reason: bool = True,
+    async_mode: bool = True,
+    **kwargs: Any,
+) -> Any:
+    """
+    DeepEval 메트릭 인스턴스 생성.
+
+    Args:
+        metric_name: 메트릭 이름 (answer_relevancy, faithfulness 등)
+        model: LLM 모델
+        threshold: 통과 임계값
+        include_reason: 평가 이유 포함 여부
+        async_mode: 비동기 모드 사용
+        **kwargs: 메트릭별 추가 파라미터
+
+    Returns:
+        DeepEval Metric 인스턴스
+
+    Raises:
+        ImportError: deepeval 미설치 시
+        ValueError: 알 수 없는 메트릭 이름
+    """
+    metric_map = get_deepeval_metric_map()
+
+    if metric_name not in metric_map:
+        raise ValueError(f"Unknown metric: {metric_name}. Available: {list(metric_map.keys())}")
+
+    metric_class = metric_map[metric_name]
+
+    return metric_class(
+        model=model,
+        threshold=threshold,
+        include_reason=include_reason,
+        async_mode=async_mode,
+        **kwargs,
+    )