LLM API Benchmark MCP Server

import numpy as np import math def calculate_statistics(data): if not data: return { "total": 0.0, "avg": 0.0, "distribution": { "max": 0.0, "p50": 0.0, "p10": 0.0, "p1": 0.0, "min": 0.0 } } data_np = np.array(data) sorted_data = np.sort(data_np) n = len(sorted_data) # Calculate p1 (actual value from sorted data) p1_index = min(n - 1, math.ceil(0.01 * n) - 1) if n > 0 else 0 p1_actual = float(sorted_data[p1_index]) if n > 0 else 0.0 # Calculate p10 (actual value from sorted data) p10_index = min(n - 1, math.ceil(0.10 * n) - 1) if n > 0 else 0 p10_actual = float(sorted_data[p10_index]) if n > 0 else 0.0 # Calculate p50 (median) - using 'lower' interpolation equivalent for actual value p50_index = math.floor(0.50 * (n - 1)) p50_actual = float(sorted_data[p50_index]) return { "total": float(np.sum(data_np)), "avg": float(np.mean(data_np)), "distribution": { "max": float(np.max(data_np)), "p50": p50_actual, "p10": p10_actual, "p1": p1_actual, "min": float(np.min(data_np)) } } def calculate_ttft_statistics(data): if not data: return { "avg": 0.0, "distribution": { "min": 0.0, "p50": 0.0, "p90": 0.0, "p99": 0.0, "max": 0.0 } } data_np = np.array(data) sorted_data = np.sort(data_np) n = len(sorted_data) # Calculate p50 (median) - using 'lower' interpolation equivalent for actual value p50_index = math.floor(0.50 * (n - 1)) p50_actual = float(sorted_data[p50_index]) # Calculate p90 (actual value from sorted data) p90_index = min(n - 1, math.ceil(0.90 * n) - 1) if n > 0 else 0 p90_actual = float(sorted_data[p90_index]) if n > 0 else 0.0 # Calculate p99 (actual value from sorted data) p99_index = min(n - 1, math.ceil(0.99 * n) - 1) if n > 0 else 0 p99_actual = float(sorted_data[p99_index]) if n > 0 else 0.0 return { "avg": float(np.mean(data_np)), "distribution": { "min": float(np.min(data_np)), "p50": p50_actual, "p90": p90_actual, "p99": p99_actual, "max": float(np.max(data_np)) } }