Update app.py

app.py

@@ -1,3 +1,53 @@
+# import gradio as gr
+# import faiss
+# import json
+# import numpy as np
+# from sentence_transformers import SentenceTransformer
+# from transformers import pipeline
+
+# embed_model = SentenceTransformer('all-MiniLM-L6-v2')
+# index = faiss.read_index("faiss_index.bin")  
+# with open("processed_chunks.json", "r") as f:
+#     chunks = json.load(f)
+
+
+# #weights load automatically to the Space
+# pipe = pipeline("text-generation", model="ibm-granite/granite-3.1-2b-instruct")
+# # pipe = pipeline("text-generation", model="ibm-granite/granite-3.3-8b-instruct")
+
+
+# def ask_hr_bot(question):
+#     # Retrieval
+#     query_vec = embed_model.encode([question])
+#     distances, indices = index.search(np.array(query_vec).astype('float32'), k=1)
+    
+#     # out-of-scope questions
+#     if distances[0][0] > 1.5:
+#         return "I could not find information on this in the HR documents.", "N/A"
+
+#     # Context Retrieval
+#     relevant_chunk = chunks[indices[0][0]]
+#     context = relevant_chunk['text']
+#     source = relevant_chunk['doc_name']
+
+#     # Grounded Generation
+#     prompt = f"Context: {context}\nQuestion: {question}\nAnswer only from context. Cite source: {source}\nAnswer:"
+#     res = pipe(prompt, max_new_tokens=60, do_sample=False)
+    
+#     answer = res[0]['generated_text'].split("Answer:")[-1].strip()
+#     return answer, source
+
+
+# interface = gr.Interface(
+#     fn=ask_hr_bot,
+#     inputs=gr.Textbox(label="Ask an HR Question"),
+#     outputs=[gr.Textbox(label="Bot Answer"), gr.Textbox(label="Source Used")],
+#     title="HR Knowledge Assistant",
+#     description="Enterprise RAG Prototype using IBM Granite Instruct Family."
+# )
+
+# interface.launch()
+
 import gradio as gr
 import faiss
 import json
@@ -5,45 +55,58 @@ import numpy as np
 from sentence_transformers import SentenceTransformer
 from transformers import pipeline
 
+
 embed_model = SentenceTransformer('all-MiniLM-L6-v2')
-index = faiss.read_index("faiss_index.bin")  
+index = faiss.read_index("faiss_index.bin")
 with open("processed_chunks.json", "r") as f:
     chunks = json.load(f)
 
 
-#weights load automatically to the Space
-# pipe = pipeline("text-generation", model="ibm-granite/granite-3.1-2b-instruct")
-pipe = pipeline("text-generation", model="ibm-granite/granite-3.3-8b-instruct")
+MODELS = {
+    "Granite 3.1 2B": "ibm-granite/granite-3.1-2b-instruct",
+    "Llama 3.2 1B": "meta-llama/Llama-3.2-1B-Instruct",
+    "Phi 3.5 Mini": "microsoft/Phi-3.5-mini-instruct",
+    "Gemma 2 2B": "google/gemma-2-2b-it"
+}
+
 
+pipes = {name: pipeline("text-generation", model=id, device_map="auto") 
+         for name, id in MODELS.items()}
 
-def ask_hr_bot(question):
-    # Retrieval
+def compare_hr_bots(question):
+    # Retrieval (Same for all models to ensure fair comparison)
     query_vec = embed_model.encode([question])
     distances, indices = index.search(np.array(query_vec).astype('float32'), k=1)
     
-    # out-of-scope questions
     if distances[0][0] > 1.5:
-        return "I could not find information on this in the HR documents.", "N/A"
+        return ["Out of scope"] * 4, "N/A"
 
-    # Context Retrieval
     relevant_chunk = chunks[indices[0][0]]
     context = relevant_chunk['text']
     source = relevant_chunk['doc_name']
-
-    # Grounded Generation
     prompt = f"Context: {context}\nQuestion: {question}\nAnswer only from context. Cite source: {source}\nAnswer:"
-    res = pipe(prompt, max_new_tokens=60, do_sample=False)
+
+    answers = []
+    for name in MODELS.keys():
+        res = pipes[name](prompt, max_new_tokens=60, do_sample=False)
+        ans = res[0]['generated_text'].split("Answer:")[-1].strip()
+        answers.append(ans)
     
-    answer = res[0]['generated_text'].split("Answer:")[-1].strip()
-    return answer, source
+    return answers[0], answers[1], answers[2], answers[3], source
 
 
 interface = gr.Interface(
-    fn=ask_hr_bot,
+    fn=compare_hr_bots,
     inputs=gr.Textbox(label="Ask an HR Question"),
-    outputs=[gr.Textbox(label="Bot Answer"), gr.Textbox(label="Source Used")],
-    title="HR Knowledge Assistant",
-    description="Enterprise RAG Prototype using IBM Granite Instruct Family."
+    outputs=[
+        gr.Textbox(label="IBM Granite 3.1"),
+        gr.Textbox(label="Meta Llama 3.2"),
+        gr.Textbox(label="Microsoft Phi 3.5"),
+        gr.Textbox(label="Google Gemma 2"),
+        gr.Textbox(label="Source Used")
+    ],
+    title="RAG Model Benchmarking: HR Assistant",
+    description="Comparing grounding quality across 4 open-source LLMs using ADU HR Policies."
 )
 
 interface.launch()