CPU Embedding Benchmark Tool: Automated Performance Comparison Across 8 CPUs

# cpu_embedding_bench.py

import os, time, json, platform
from sentence_transformers import SentenceTransformer

MODELS = [
    "all-MiniLM-L6-v2",
    "BAAI/bge-base-en-v1.5",
    "intfloat/multilingual-e5-large",
]

sentences = [
    f"This is test sentence number {i} for "
    f"embedding benchmark."
    for i in range(1000)
]

def benchmark_model(model_name, batch_size=32, runs=3):
    model = SentenceTransformer(model_name)
    model.encode(sentences[:10], batch_size=batch_size)  # warmup

    times = []
    for _ in range(runs):
        start = time.perf_counter()
        model.encode(sentences, batch_size=batch_size)
        elapsed = time.perf_counter() - start
        times.append(elapsed)

    median_time = sorted(times)[len(times) // 2]
    return {
        "model": model_name,
        "median_seconds": round(median_time, 2),
        "sentences_per_sec": round(len(sentences) / median_time, 1),
    }

results = {
    "cpu": platform.processor(),
    "cores": os.cpu_count(),
    "benchmarks": [benchmark_model(m) for m in MODELS],
}
print(json.dumps(results, indent=2))

# Running the benchmark

pip install sentence-transformers torch

# Run and save results
python cpu_embedding_bench.py > results.json

# Limit to specific core count
OMP_NUM_THREADS=4 python cpu_embedding_bench.py

# remote-bench.sh — run benchmarks on multiple servers

#!/bin/bash
SERVERS=("server-a" "server-b" "server-c" "minipc-a")
SCRIPT_PATH="/opt/bench/cpu_embedding_bench.py"
OUTPUT_DIR="./results"

mkdir -p "$OUTPUT_DIR"

for server in "${SERVERS[@]}"; do
  echo "[$server] Starting benchmark..."
  ssh "$server" "python3 $SCRIPT_PATH" \
    > "$OUTPUT_DIR/$server.json" 2>/dev/null

  if [ $? -eq 0 ]; then
    echo "[$server] Complete"
  else
    echo "[$server] Failed"
  fi
done
echo "All benchmarks complete. Results: $OUTPUT_DIR/"

# Single server output example (server-a.json)

{
  "cpu": "16-core desktop CPU",
  "cores": 32,
  "benchmarks": [
    { "model": "all-MiniLM-L6-v2",
      "median_seconds": 1.18,
      "sentences_per_sec": 847.3 },
    { "model": "BAAI/bge-base-en-v1.5",
      "median_seconds": 3.42,
      "sentences_per_sec": 292.4 },
    { "model": "intfloat/multilingual-e5-large",
      "median_seconds": 8.75,
      "sentences_per_sec": 114.3 }
  ]
}

# merge_results.py — combine multi-server results

import json, glob

results = []
for path in sorted(glob.glob("results/*.json")):
    with open(path) as f:
        data = json.load(f)
        data["server"] = path.split("/")[-1].replace(".json", "")
        results.append(data)

for model_idx in range(len(results[0]["benchmarks"])):
    model_name = results[0]["benchmarks"][model_idx]["model"]
    print(f"\n=== {model_name} ===")
    ranked = sorted(
        results,
        key=lambda r: r["benchmarks"][model_idx]["sentences_per_sec"],
        reverse=True,
    )
    for r in ranked:
        bench = r["benchmarks"][model_idx]
        print(f"  {r['cpu']:30s}  {bench['sentences_per_sec']:8.1f} sent/s")