updated models

Dockerfile

@@ -24,7 +24,7 @@ RUN pip install --no-cache-dir "youtube-transcript-api>=1.2.4"
 
 COPY --chown=appuser:appuser . .
 
-RUN mkdir -p /app/models && python3 -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='BAAI/bge-small-en-v1.5', local_dir='/app/models/bge-384'); snapshot_download(repo_id='BAAI/bge-base-en-v1.5', local_dir='/app/models/bge-768'); snapshot_download(repo_id='BAAI/bge-large-en-v1.5', local_dir='/app/models/bge-1024')" && chown -R appuser:appuser /app/models
+RUN mkdir -p /app/models && python3 -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='ibm-granite/granite-embedding-small-english-r2', local_dir='/app/models/bge-384'); snapshot_download(repo_id='nomic-ai/modernbert-embed-base', local_dir='/app/models/bge-768'); snapshot_download(repo_id='lightonai/modernbert-embed-large', local_dir='/app/models/bge-1024')" && chown -R appuser:appuser /app/models
 
 RUN mkdir -p /app/logs && \
     chown -R appuser:appuser /app/logs

app/api/v1/embeddings.py

@@ -3,12 +3,13 @@ from __future__ import annotations
 import asyncio
 import concurrent.futures
 import os
+import time
 
 from fastapi import APIRouter, Depends, HTTPException
 
 from app.api.deps import require_auth, get_embeddings_service
 from app.core.logger import get_logger
-from app.models.schemas import EmbeddingRequest, EmbeddingResponse
+from app.models.schemas import EmbeddingItem, EmbeddingRequest, EmbeddingResponse
 from app.services.embeddings_service import EmbeddingService
 
 router = APIRouter()
@@ -27,7 +28,7 @@ async def create_embeddings(
     token: str = Depends(require_auth),
     embedding_service: EmbeddingService = Depends(get_embeddings_service),
 ) -> EmbeddingResponse:
-    _logger.info("Embedding request: dim=%s, content_len=%s", body.dimension, len(body.content))
+    _logger.info("Embedding request: dim=%s, items=%s", body.dimension, len(body.content))
 
     if not embedding_service.is_loaded(body.dimension):
         _logger.error("Model dim=%s not loaded. Loaded: %s", body.dimension, embedding_service.loaded_dimensions)
@@ -39,24 +40,52 @@ async def create_embeddings(
             },
         )
 
+    start = time.perf_counter()
     try:
         loop = asyncio.get_running_loop()
-        embeddings = await loop.run_in_executor(
+        vectors = await loop.run_in_executor(
             _thread_pool,
             embedding_service.generate_embedding,
             body.content,
             body.dimension,
         )
-
-        _logger.info("Embedding success: dim=%s, vector_len=%s", body.dimension, len(embeddings))
+    except Exception as exc:
+        elapsed = (time.perf_counter() - start) * 1000
+        _logger.error("Embedding error: %s", exc)
         return EmbeddingResponse(
+            success=False,
+            time_ms=round(elapsed, 3),
+            success_count=0,
+            failed_count=len(body.content),
+            error_message=str(exc),
+            results=[
+                EmbeddingItem(
+                    success=False,
+                    time_ms=0,
+                    error_message=str(exc),
+                )
+                for _ in body.content
+            ],
+        )
+
+    total_ms = (time.perf_counter() - start) * 1000
+    per_item_ms = total_ms / len(body.content)
+
+    results = [
+        EmbeddingItem(
             success=True,
-            embeddings=embeddings,
+            time_ms=round(per_item_ms, 3),
+            embeddings=vec,
             dimension=body.dimension,
         )
-    except Exception as exc:
-        _logger.error("Embedding error: %s", exc)
-        raise HTTPException(
-            status_code=500,
-            detail={"success": False, "message": str(exc)},
-        )
+        for vec in vectors
+    ]
+
+    _logger.info("Embedding success: dim=%s, items=%s, total_ms=%s", body.dimension, len(results), round(total_ms, 3))
+    return EmbeddingResponse(
+        success=True,
+        time_ms=round(total_ms, 3),
+        success_count=len(results),
+        failed_count=0,
+        results=results,
+    )

app/models/__init__.py

@@ -7,6 +7,7 @@ from app.models.schemas import (
     BatchUrlRequest,
     ConversionMetadata,
     ConversionResponse,
+    EmbeddingItem,
     EmbeddingRequest,
     EmbeddingResponse,
     HealthResponse,
@@ -21,6 +22,7 @@ __all__ = [
     "ConversionResult",
     "ConversionMetadata",
     "ConversionResponse",
+    "EmbeddingItem",
     "EmbeddingRequest",
     "EmbeddingResponse",
     "UrlRequest",

app/models/schemas.py

@@ -224,12 +224,23 @@ class DatabaseQueryResponse(BaseModel):
     error: Optional[DatabaseQueryError] = None
 
 
+class EmbeddingItem(BaseModel):
+    success: bool
+    time_ms: float
+    embeddings: List[float] = Field(default_factory=list)
+    dimension: int = 0
+    error_message: Optional[str] = None
+
+
 class EmbeddingRequest(BaseModel):
-    content: str = Field(..., min_length=1, description="Text to embed")
+    content: List[str] = Field(..., min_length=1, max_length=10, description="Array of text strings to embed (max 10)")
     dimension: int = Field(default=384, ge=384, le=1024, description="Target embedding dimension (384, 768, or 1024)")
 
 
 class EmbeddingResponse(BaseModel):
     success: bool
-    embeddings: List[float]
-    dimension: int
+    time_ms: float
+    success_count: int
+    failed_count: int
+    error_message: Optional[str] = None
+    results: List[EmbeddingItem]

app/services/embeddings_service.py

@@ -10,9 +10,9 @@ from sentence_transformers import SentenceTransformer
 _logger = logging.getLogger(__name__)
 
 _MODEL_MAP: Dict[int, str] = {
-    384: "BAAI/bge-small-en-v1.5",
-    768: "BAAI/bge-base-en-v1.5",
-    1024: "BAAI/bge-large-en-v1.5",
+    384: "ibm-granite/granite-embedding-small-english-r2",
+    768: "nomic-ai/modernbert-embed-base",
+    1024: "lightonai/modernbert-embed-large",
 }
 
 
@@ -53,7 +53,7 @@ class EmbeddingService:
         for dim in _MODEL_MAP:
             self.load_model(dim)
 
-    def generate_embedding(self, text: str, dimension: int) -> List[float]:
+    def generate_embedding(self, text: List[str], dimension: int) -> List[List[float]]:
         if dimension not in self._models:
             raise ValueError(f"Model for dimension {dimension} not loaded")
         model = self._models[dimension]

requirements.txt

@@ -11,7 +11,9 @@ onnxruntime>=1.18.0
 pillow>=10.0.0
 pypdfium2>=4.30.0
 pandas>=2.0.0
-sentence-transformers==3.3.1
+sentence-transformers==5.6.0
+transformers==5.12.1
+torch==2.12.1
 spacy>=3.7.0
 phonenumbers>=8.13.0