asmhatreandClaude Sonnet 5.5
c97bcf9adf
Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_017ic4b4VQVEJWetTT1dYzyJ
35 lines
1.9 KiB
Bash
Executable File
35 lines
1.9 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# bench.sh <label> <base-url> <model> <container> [extra bench args]
|
|
# Runs the Spring AI benchmark against one OpenAI-compatible server and samples the container's
|
|
# memory with `docker stats` once a second while it runs. Writes output/<label>.txt.
|
|
# Docker Model Runner : bench.sh dmr http://localhost:12434/engines/v1 hf.co/qwen/qwen2.5-0.5b-instruct-gguf:Q4_K_M dmr
|
|
# Ollama : bench.sh ollama http://localhost:11434/v1 qwen2.5:0.5b ollama
|
|
set -euo pipefail
|
|
cd "$(dirname "$0")/.."
|
|
LABEL=$1; URL=$2; MODEL=$3; CTR=$4; shift 4
|
|
[ -f target/docker-model-runner-1.0.0.jar ] || mvn -B -q -T1 package -DskipTests
|
|
mkdir -p output
|
|
MEM=$(mktemp)
|
|
( while true; do docker stats --no-stream --format '{{.MemUsage}}' "$CTR" | cut -d/ -f1 >> "$MEM"; sleep 1; done ) &
|
|
SAMPLER=$!
|
|
trap 'kill $SAMPLER 2>/dev/null || true' EXIT
|
|
echo "idle container memory before: $(docker stats --no-stream --format '{{.MemUsage}}' "$CTR" | cut -d/ -f1)" > "output/$LABEL.mem.tmp"
|
|
java -jar target/docker-model-runner-1.0.0.jar --bench.run=true --bench.label="$LABEL" \
|
|
--bench.out-file="output/$LABEL.txt" --spring.ai.openai.base-url="$URL" \
|
|
--spring.ai.openai.chat.options.model="$MODEL" "$@" > "output/$LABEL.log" 2>&1 || { tail -20 "output/$LABEL.log"; exit 1; }
|
|
kill $SAMPLER 2>/dev/null || true
|
|
python3 - "$MEM" "output/$LABEL.mem.tmp" "output/$LABEL.txt" <<'P'
|
|
import sys,re
|
|
def mb(s):
|
|
s=s.strip(); m=re.match(r'([\d.]+)(\w+)',s)
|
|
if not m: return None
|
|
v=float(m[1]); u=m[2]
|
|
return v*{'B':1/1048576,'KiB':1/1024,'MiB':1,'GiB':1024}[u]
|
|
vals=[x for x in (mb(l) for l in open(sys.argv[1])) if x is not None]
|
|
idle=open(sys.argv[2]).read().strip()
|
|
with open(sys.argv[3],'a') as f:
|
|
f.write(f"\nmemory ({idle})\ncontainer memory during run: peak {max(vals):.0f} MiB, median {sorted(vals)[len(vals)//2]:.0f} MiB, {len(vals)} samples\n")
|
|
P
|
|
rm -f "output/$LABEL.mem.tmp" "output/$LABEL.log"
|
|
cat "output/$LABEL.txt"
|