oliver-aizip commited on
Commit
c77435d
·
1 Parent(s): c3e4586

add requirements

Browse files
Files changed (2) hide show
  1. app.py +102 -0
  2. requirements.txt +8 -0
app.py CHANGED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ import random
3
+ import json
4
+ import os
5
+ from datetime import datetime
6
+
7
+ # This would be replaced with your actual SLM integration
8
+ def generate_response(query, context, model_name):
9
+ """Placeholder function to generate response from an SLM"""
10
+ return f"This is a placeholder response from {model_name} based on query: {query} and context: {context}"
11
+
12
+ def save_evaluation(query, context, model_a, model_b, response_a, response_b, preference):
13
+ """Save evaluation results to a JSON file"""
14
+ timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
15
+ evaluation = {
16
+ "timestamp": timestamp,
17
+ "query": query,
18
+ "context": context,
19
+ "models": {
20
+ "left": model_a,
21
+ "right": model_b
22
+ },
23
+ "responses": {
24
+ "left": response_a,
25
+ "right": response_b
26
+ },
27
+ "preference": preference
28
+ }
29
+
30
+ # Create directory if it doesn't exist
31
+ os.makedirs("evaluations", exist_ok=True)
32
+
33
+ # Save to a file
34
+ with open(f"evaluations/eval_{timestamp.replace(' ', '_').replace(':', '-')}.json", "w") as f:
35
+ json.dump(evaluation, f, indent=2)
36
+
37
+ return "Evaluation saved successfully!"
38
+
39
+ def process_query(query, context, model_a="SLM-A", model_b="SLM-B"):
40
+ """Process query and generate responses from two models"""
41
+ # Generate responses
42
+ response_a = generate_response(query, context, model_a)
43
+ response_b = generate_response(query, context, model_b)
44
+
45
+ # Randomly swap to avoid position bias
46
+ if random.random() > 0.5:
47
+ return response_a, response_b, model_a, model_b
48
+ else:
49
+ return response_b, response_a, model_b, model_a
50
+
51
+ def submit_evaluation(query, context, response_left, response_right, preference, model_left, model_right):
52
+ """Submit and save the evaluation"""
53
+ if not preference:
54
+ return "Please select a preference before submitting."
55
+
56
+ save_evaluation(query, context, model_left, model_right, response_left, response_right, preference)
57
+ return "Thank you for your evaluation!"
58
+
59
+ with gr.Blocks(title="SLM-RAG Arena") as app:
60
+ gr.Markdown("# SLM-RAG Arena")
61
+ gr.Markdown("Compare responses from different models for RAG tasks.")
62
+
63
+ with gr.Row():
64
+ with gr.Column():
65
+ query_input = gr.Textbox(label="Query", placeholder="Enter your query here...")
66
+ context_input = gr.Textbox(label="Context", placeholder="Enter context information here...", lines=5)
67
+ generate_btn = gr.Button("Generate Responses")
68
+
69
+ # Hidden state variables
70
+ model_left = gr.State("")
71
+ model_right = gr.State("")
72
+
73
+ with gr.Row():
74
+ with gr.Column():
75
+ gr.Markdown("### Response A")
76
+ response_left = gr.Textbox(label="", lines=10, interactive=False)
77
+ with gr.Column():
78
+ gr.Markdown("### Response B")
79
+ response_right = gr.Textbox(label="", lines=10, interactive=False)
80
+
81
+ with gr.Row():
82
+ preference = gr.Radio(
83
+ choices=["Prefer Left", "Tie", "Prefer Right", "Neither"],
84
+ label="Which response do you prefer?"
85
+ )
86
+
87
+ submit_btn = gr.Button("Submit Evaluation")
88
+ result = gr.Textbox(label="Result")
89
+
90
+ generate_btn.click(
91
+ process_query,
92
+ inputs=[query_input, context_input],
93
+ outputs=[response_left, response_right, model_left, model_right]
94
+ )
95
+
96
+ submit_btn.click(
97
+ submit_evaluation,
98
+ inputs=[query_input, context_input, response_left, response_right, preference, model_left, model_right],
99
+ outputs=[result]
100
+ )
101
+
102
+ app.launch()
requirements.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ python>=3.10
2
+ transformers>=4.51.0
3
+ pandas>=2.2.3
4
+ accelerate>=1.6.0
5
+ numpy==1.26.4
6
+ openai>=1.60.2
7
+ torch>=2.5.1
8
+ tqdm==4.67.1