Create app.py

From codellama70B

app.py

@@ -0,0 +1,43 @@
+import gradio as gr
+import pandas as pd
+import requests
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+from rag_embeddings import RagRetriever
+
+# Load the RAG model
+rag_retriever = RagRetriever("rag_sequence")
+
+# Load the Mistral model
+tokenizer = AutoTokenizer.from_pretrained("EleutherAI/mistral-small-12L-4H-768d-albert")
+model = AutoModelForSeq2SeqLM.from_pretrained("EleutherAI/mistral-small-12L-4H-768d-albert")
+
+# Load the tabular data
+data = pd.read_csv("data.csv")
+#ADD DATASETS HERE
+
+def langchain(user_prompt, master_prompt):
+    # Retrieve data chunks using RAG embeddings
+    retrieved_data = rag_retriever.retrieve(user_prompt, data, num_results=5)
+
+    # Connect retrieved data chunks to user prompt and master prompt
+    input_text = user_prompt + " " + master_prompt + " " + " ".join(retrieved_data)
+
+    # Generate response using Mistral model
+    input_ids = tokenizer(input_text, return_tensors="pt").input_ids
+    generated_ids = model.generate(input_ids)
+    response = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
+
+    return response
+
+iface = gr.Interface(
+    fn=langchain,
+    inputs=["text", "text"],
+    outputs="text",
+    title="LangChain App",
+    description="A Gradio app that retrieves specific datachunks using RAG embeding, from tabular csv data, and then connects those into the user prompt and the master prompt and then feed them into a Mistral model called from Hugging Face ran locally, then returns the response to the user via the gradio app GUI.",
+)
+
+iface.launch()
+
+iface.launch(share=True)
+