Fix HF Spaces compatibility (models + langchain imports)

app.py

@@ -36,4 +36,4 @@ with gr.Blocks(title="Tech Explainer RAG") as demo:
 
 
 if __name__ == "__main__":
-    demo.launch(share=True)
+    demo.launch()

config.py

@@ -1,3 +1,3 @@
 EMBEDDING_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
-PRIMARY_LLM = "mistralai/Mistral-7B-Instruct-v0.2"
-FALLBACK_LLM = "microsoft/phi-2"
+PRIMARY_LLM = "google/gemma-2b-it"
+FALLBACK_LLM = "microsoft/phi-2"

rag/llm.py

@@ -7,7 +7,7 @@ def load_model():
     try:
         tokenizer = AutoTokenizer.from_pretrained(PRIMARY_LLM)
         model = AutoModelForCausalLM.from_pretrained(
-            PRIMARY_LLM, device_map="auto", load_in_8bit=True
+            PRIMARY_LLM, device_map="auto"
         )
     except Exception:
         tokenizer = AutoTokenizer.from_pretrained(FALLBACK_LLM, trust_remote_code=True)