Update backend/settings.py

backend/settings.py

@@ -2,10 +2,7 @@ from pathlib import Path
 from pydantic import BaseModel
 import os
 
-LLM_N_GPU_LAYERS: int = 0
-
 class Settings(BaseModel):
-
     ROOT: Path = Path(".")
     DATA_DIR: Path = ROOT / "backend" / "data"
     RAW_DIR: Path = DATA_DIR / "raw"
@@ -14,14 +11,17 @@ class Settings(BaseModel):
     MODEL_DIR: Path = ROOT / "backend" / "models"
 
 
-    EMBEDDING_MODEL_ID: str = "BAAI/bge-m3"
-    RERANK_MODEL_ID: str = "BAAI/bge-reranker-v2-m3"  
-    LLM_REPO_ID: str = "Qwen/Qwen2.5-3B-Instruct-GGUF"
-    LLM_FILENAME_PATTERN: str = "Qwen2.5-3B-Instruct-Q4_K_M.gguf"
+    EMBEDDING_MODEL_ID: str = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2" 
+
+
+    RERANK_MODEL_ID: str = "BAAI/bge-reranker-v2-m3"
+
 
+    LLM_REPO_ID: str = "Qwen/Qwen2.5-1.5B-Instruct-GGUF"
+    LLM_FILENAME_PATTERN: str = "Qwen2.5-1.5B-Instruct-Q4_K_M.gguf"
 
-    LLM_PATH: Path = MODEL_DIR / "qwen2.5-3b-gguf" / "Qwen2.5-3B-Instruct-Q4_K_M.gguf"
 
+    LLM_PATH: Path = MODEL_DIR / "qwen2.5-1.5b-gguf" / "Qwen2.5-1.5B-Instruct-Q4_K_M.gguf"
 
     N_CHARS_CHUNK: int = 2000
     N_CHARS_OVERLAP: int = 300
@@ -29,16 +29,14 @@ class Settings(BaseModel):
     TOP_K_RERANK: int = 3
 
 
-    LLM_N_CTX: int = 4096
-    LLM_N_GPU_LAYERS: int = 0  
-    
-    HF_TOKEN: str | None = (
-    os.getenv("study")    
-    or os.getenv("HF_TOKEN")
-    or os.getenv("HUGGINGFACE_HUB_TOKEN")
-)
-
-settings = Settings()
+    LLM_N_CTX: int = 3072 
+    LLM_N_GPU_LAYERS: int = 0 
 
 
+    HF_TOKEN: str | None = (
+        os.getenv("HF_TOKEN")
+        or os.getenv("HUGGINGFACE_HUB_TOKEN")
+        or os.getenv("study")
+    )
 
+settings = Settings()