1
0

run-all-perf.sh 549 B

12345678910111213141516171819202122232425262728293031323334
  1. #!/bin/bash
  2. qnt=(f16 q8_0 q6_k q5_k q5_1 q5_0 q4_k q4_1 q4_0 q3_k q2_k)
  3. args="-ngl 999 -n 64 -p 512"
  4. if [ -z "$1" ]; then
  5. echo "usage: $0 <model> [qnt] [args]"
  6. echo "default: $0 <model> \"${qnt[@]}\" \"${args}\""
  7. exit 1
  8. fi
  9. if [ ! -z "$2" ]; then
  10. qnt=($2)
  11. fi
  12. if [ ! -z "$3" ]; then
  13. args="$3"
  14. fi
  15. model="$1"
  16. out="../tmp/results-${model}"
  17. set -o pipefail
  18. set -e
  19. mkdir -p ${out}
  20. mstr=""
  21. for q in ${qnt[@]}; do
  22. mstr="${mstr} -m ../models/${model}/ggml-model-${q}.gguf"
  23. done
  24. ./bin/llama-bench ${mstr} ${args} 2> /dev/null