interactive.py

# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
# 
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
#  * Redistributions of source code must retain the above copyright
#    notice, this list of conditions and the following disclaimer.
#  * Redistributions in binary form must reproduce the above copyright
#    notice, this list of conditions and the following disclaimer in the
#    documentation and/or other materials provided with the distribution.
#  * Neither the name of NVIDIA CORPORATION nor the names of its
#    contributors may be used to endorse or promote products derived
#    from this software without specific prior written permission.
# 
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

"""
This script is used to compare the estimated latency of different models interactively.
Outputs table is displayed in the terminal.
"""


import latency_model

def format_time(seconds):
    if seconds < 0.001:
        return f"{seconds*1000000:.1f}μs"
    elif seconds < 1:
        return f"{seconds*1000:.1f}ms"
    else:
        return f"{seconds:.3f}s"

def print_model_comparison(input_length, output_length):
    results = {}
    for model_name in latency_model.prefill_coeffs.keys():
        prefill_lat, decode_lat, total_lat = latency_model.total_latency_model(model_name, input_length, output_length)
        results[model_name] = {
            'prefill': prefill_lat,
            'decode': decode_lat, 
            'total': total_lat
        }
    
    print("=" * 90)
    print(f"Latency Model Comparison")
    print(f"Input: {input_length:,} tokens | Output: {output_length:,} tokens")
    print("=" * 90)
    
    print(f"{'Model':<20} {'Prefill':<12} {'Decode':<12} {'Total':<12} {'Ratio':<8} {'Speed':<10}")
    print("-" * 90)
    
    sorted_models = sorted(results.items(), key=lambda x: x[1]['total'])
    fastest_total = sorted_models[0][1]['total']
    
    for model_name, latencies in sorted_models:
        prefill_str = format_time(latencies['prefill'])
        decode_str = format_time(latencies['decode'])
        total_str = format_time(latencies['total'])
        
        ratio = latencies['decode'] / latencies['prefill'] if latencies['prefill'] > 0 else 0
        speed_multiplier = latencies['total'] / fastest_total
        speed_str = f"{speed_multiplier:.1f}x" if speed_multiplier > 1 else "baseline"
        
        print(f"{model_name:<20} {prefill_str:<12} {decode_str:<12} {total_str:<12} {ratio:<8.1f} {speed_str:<10}")
    
    print("-" * 90)
    
    fastest_model = sorted_models[0][0]
    slowest_model = sorted_models[-1][0]
    
    print("\n Insights:")
    print(f"   Fastest: {fastest_model}")
    print(f"   Slowest: {slowest_model}")
    
    total_tokens = input_length + output_length
    print(f"\n📈 TOKEN PROCESSING RATES:")
    for model_name, latencies in sorted_models:
        tokens_per_sec = total_tokens / latencies['total']
        print(f"   {model_name}: {tokens_per_sec:,.0f} tokens/sec")
    
    print(f"\n⚡ LATENCY BREAKDOWN:")
    for model_name, latencies in results.items():
        prefill_pct = (latencies['prefill'] / latencies['total']) * 100
        decode_pct = (latencies['decode'] / latencies['total']) * 100
        print(f"   {model_name}: Prefill {prefill_pct:.1f}% | Decode {decode_pct:.1f}%")

def test_multiple_scenarios():
    scenarios = [
        {"name": "Short Query", "input": 100, "output": 50},
        {"name": "Medium Query", "input": 500, "output": 200}, 
        {"name": "Long Query", "input": 2000, "output": 500},
        {"name": "Very Long Query", "input": 8000, "output": 1000},
    ]
    
    print("\n" + "=" * 90)
    print("Sweep")
    print("=" * 90)
    
    for scenario in scenarios:
        print(f"\n📋 {scenario['name']} ({scenario['input']} → {scenario['output']} tokens)")
        print_model_comparison(scenario['input'], scenario['output'])
        print()

def run_interactive_mode():
    input_length = 116
    output_length = 82
    
    print_model_comparison(input_length, output_length)
    
    print("\n" + "="*50)
    response = input("Sweep? (y/n): ").lower().strip()
    if response in ['y', 'yes']:
        test_multiple_scenarios()
    
    print("\n" + "="*50) 
    print("INTERACTIVE MODE")
    print("="*50)
    
    while True:
        try:
            response = input("\n input,output tokens (e.g., '1000,300') or 'q' to quit: ").strip()
            if response.lower() in ['q', 'quit', 'exit']:
                break
                
            if ',' in response:
                input_tokens, output_tokens = map(int, response.split(','))
                print_model_comparison(input_tokens, output_tokens)
            else:
                print("X input,output (e.g., '1000,300')")
                
        except ValueError:
            print("X input,output (e.g., '1000,300')")
        except KeyboardInterrupt:
            break
    

if __name__ == "__main__":
    run_interactive_mode()