🔁 Switched to flan-t5-large for stable CPU generation

app.py

@@ -12,13 +12,15 @@ from langchain_community.embeddings import HuggingFaceEmbeddings
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain_core.documents import Document
 
-# Load Reasoning Model (CPU-friendly with better reasoning)
-model_id = "microsoft/phi-1_5"
+# Load Reasoning Model (CPU-friendly + stronger than base)
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
+
+model_id = "google/flan-t5-large"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
 
 reasoning_pipeline = pipeline(
-    "text-generation",
+    "text2text-generation",
     model=model,
     tokenizer=tokenizer,
     max_new_tokens=512,
@@ -26,6 +28,7 @@ reasoning_pipeline = pipeline(
     top_p=0.9
 )
 
+
 # Embedding Model
 embedding_model = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
 vectorstore = None

requirements.txt

@@ -6,6 +6,6 @@ python-docx
 gradio
 PyPDF2
 langchain
-langchain_community
 sentence-transformers
+langchain_community
 accelerate