change retrieval

core/api/server.py

@@ -32,7 +32,7 @@ from core.rag.retrieval import Retriever, RetrievalMode, get_retrieval_config
 from core.rag.generator import RAGContextBuilder, SYSTEM_PROMPT
 
 # Config
-RETRIEVAL_MODE = RetrievalMode.HYBRID_RERANK
+RETRIEVAL_MODE = RetrievalMode.HYBRID
 RETRIEVAL_CFG = get_retrieval_config()
 LLM_MODEL = os.getenv("LLM_MODEL", "qwen/qwen3-32b")
 LLM_API_BASE = "https://api.groq.com/openai/v1"

core/api/static/index.html

@@ -115,10 +115,26 @@
 
             // 2. Bảo vệ các công thức toán (MathJax) khỏi bị lỗi khi render Markdown
             let mathBlocks = [];
+            
+            // Protect block math: $$...$$ and \[...\]
             text = text.replace(/\$\$([\s\S]*?)\$\$/g, function(match) {
                 mathBlocks.push(match);
                 return `__MATH_BLOCK_${mathBlocks.length - 1}__`;
             });
+            text = text.replace(/\\\[([\s\S]*?)\\\]/g, function(match) {
+                mathBlocks.push(match);
+                return `__MATH_BLOCK_${mathBlocks.length - 1}__`;
+            });
+            
+            // Protect inline math: \(...\) and $...$
+            text = text.replace(/\\\(([\s\S]*?)\\\)/g, function(match) {
+                mathBlocks.push(match);
+                return `__MATH_BLOCK_${mathBlocks.length - 1}__`;
+            });
+            text = text.replace(/\$([^\$\n]+?)\$/g, function(match) {
+                mathBlocks.push(match);
+                return `__MATH_BLOCK_${mathBlocks.length - 1}__`;
+            });
 
             // 3. Render Markdown cơ bản
             let html = text

core/rag/retrieval.py

@@ -32,8 +32,8 @@ class RetrievalMode(str, Enum):
 class RetrievalConfig:
     rerank_api_base_url: str = "https://api.siliconflow.com/v1"
     rerank_model: str = "Qwen/Qwen3-Reranker-8B"
-    rerank_top_n: int = 10
-    initial_k: int = 25
+    rerank_top_n: int = 5
+    initial_k: int = 50
     top_k: int = 5
     vector_weight: float = 0.5
     bm25_weight: float = 0.5

requirements.txt

@@ -3,6 +3,8 @@ langchain==1.2.0
 langchain-chroma==1.1.0
 langchain-openai==1.1.6
 chromadb==1.3.7
+langchain-classic==1.0.3
+langchain_community==0.4.1
 
 # LLM & Embeddings
 groq==0.37.1
@@ -23,9 +25,12 @@ fastapi==0.115.12
 uvicorn==0.34.2
 
 # Evaluation
-ragas==0.4.2
-datasets==4.4.1
+# ragas==0.4.2
+# datasets==4.4.1
 
 # Utils
 requests==2.32.5
-huggingface-hub==0.36.0
+huggingface-hub==0.36.0
+
+# Ocr
+docling