test-embed.sh

#!/bin/bash

# Example usage: test with 50 requests (first sequentially, then in parallel)
# ./test-embed.sh 50

# Configuration
MODEL=${MODEL:-"nomic-embed-text"}  # Default model
NUM_REQUESTS=${1:-20}               # Number of requests
HOST=${HOST:-"localhost"}           # Ollama host
PORT=${PORT:-"11434"}               # Ollama port

# Create log directory
LOG_DIR="./embed_test_logs"
mkdir -p "$LOG_DIR"

# Create two distinct test texts
TEXT1="Longer text (that changes due to including appended test number): "
TEXT2="a" # short text to test if < 4 tokens is still an issue (it appears it might no longer be an issue): https://github.com/ggml-org/llama.cpp/issues/6722#issuecomment-2785033321

# High-precision time measurement function
get_time_ms() {
  echo $(date +%s.%N)
}

# Calculate duration in milliseconds
calc_duration_ms() {
  local start=$1
  local end=$2
  echo $(echo "($end - $start) * 1000" | bc | cut -d'.' -f1)
}

echo "===== Ollama Embedding Parallelism Test (Multiple Inputs) ====="
echo "Model: $MODEL"
echo "Requests: $NUM_REQUESTS (each with 2 texts)"
echo "Timestamp: $(date)"
echo "============================================================="

# Run sequential requests as baseline
echo ""
echo "Running $NUM_REQUESTS sequential embedding requests (multiple inputs)..."
SEQ_START=$(get_time_ms)

for i in $(seq 1 $NUM_REQUESTS); do
  echo "Starting sequential request $i..."
  REQ_START=$(get_time_ms)
  
  curl -s -X POST "http://$HOST:$PORT/api/embed" \
    -H "Content-Type: application/json" \
    -d "{\"model\": \"$MODEL\", \"input\": [\"$TEXT1 $i\", \"$TEXT2\"]}" \
    > "$LOG_DIR/seq_multi_$i.json"
  
  REQ_END=$(get_time_ms)
  DURATION_MS=$(calc_duration_ms $REQ_START $REQ_END)
  echo "Request $i completed in ${DURATION_MS}ms"
done

SEQ_END=$(get_time_ms)
SEQ_TOTAL_MS=$(calc_duration_ms $SEQ_START $SEQ_END)
echo "Sequential test completed in ${SEQ_TOTAL_MS}ms"

# Run parallel requests
echo ""
echo "Running $NUM_REQUESTS parallel embedding requests (multiple inputs)..."
PAR_START=$(get_time_ms)

# Start all requests in parallel
for i in $(seq 1 $NUM_REQUESTS); do
  echo "Starting parallel request $i..."
  curl -s -X POST "http://$HOST:$PORT/api/embed" \
    -H "Content-Type: application/json" \
    -d "{\"model\": \"$MODEL\", \"input\": [\"$TEXT1 $i\", \"$TEXT2\"]}" \
    > "$LOG_DIR/par_multi_$i.json" &
done

# Wait for all background processes to complete
wait

PAR_END=$(get_time_ms)
PAR_TOTAL_MS=$(calc_duration_ms $PAR_START $PAR_END)
echo "Parallel test completed in ${PAR_TOTAL_MS}ms"

# Calculate speedup
if [ "$PAR_TOTAL_MS" -gt 0 ]; then
  SPEEDUP=$(echo "scale=2; $SEQ_TOTAL_MS / $PAR_TOTAL_MS" | bc)
else
  SPEEDUP="N/A"
fi

# Report results
echo ""
echo "===== RESULTS SUMMARY ====="
echo "Sequential execution time: ${SEQ_TOTAL_MS}ms"
echo "Parallel execution time:   ${PAR_TOTAL_MS}ms"
echo "Speedup:                   ${SPEEDUP}x"
echo "============================="

if (( $(echo "$SPEEDUP > 1.2" | bc -l 2>/dev/null) )); then
  echo "SUCCESS: Parallelism is working! Significant speedup detected."
elif (( $(echo "$SPEEDUP >= 0.9" | bc -l 2>/dev/null) )); then
  echo "PARTIAL: Limited parallelism detected. Some speedup but not optimal."
else
  echo "FAILED: No parallelism detected. Parallel execution was slower than sequential."
fi

echo ""
echo "Log files saved to: $LOG_DIR"

# Validate results by checking response format
echo ""
echo "Validating responses..."
SAMPLE_FILE=$(find $LOG_DIR -name "par_multi_1.json" -o -name "seq_multi_1.json" | head -1)

if [ -f "$SAMPLE_FILE" ]; then
  # Check if we have embeddings array in the response
  EMBEDDING_COUNT=$(grep -o "\"embeddings\":" "$SAMPLE_FILE" | wc -l)
  if [ "$EMBEDDING_COUNT" -gt 0 ]; then
    echo "Response format looks good, found embeddings array."
    
    # Check if we have 2 embeddings per request
    ARRAYS=$(grep -o "\\[\\[" "$SAMPLE_FILE" | wc -l)
    if [ "$ARRAYS" -ge 2 ]; then
      echo "SUCCESS: Found multiple embeddings in the response (one for each input text)."
    else
      echo "WARNING: Expected multiple embeddings arrays but found fewer than expected."
    fi
  else
    echo "ERROR: Response format issue - no embeddings found in response."
  fi
else
  echo "WARNING: Could not find any response files to validate."
fi

# Check if any responses contain errors
grep -l "error" $LOG_DIR/*.json > /dev/null 2>&1
if [ $? -eq 0 ]; then
  echo "WARNING: Some responses contain errors! Check log files."
  grep -l "error" $LOG_DIR/*.json | while read file; do
    echo "Error in $file:"
    cat "$file" | grep -A 2 "error"
  done
else
  echo "No errors found in responses."
fi