scripts/rtx-docker/rtx_fix_serving.sh

#!/usr/bin/env bash
# Re-run dos 3 serving zerados (14B-BF16, Coder-7B-BF16, 70B-AWQ).
# CAUSA DO ZERO: --base-url tinha /v1 → vllm bench vira /v1/v1/completions → 404.
# FIX: base-url SEM /v1 (o vllm bench anexa /v1/completions sozinho).
set -u
IMG="nvcr.io/nvidia/vllm:26.03.post1-py3"
TAG="NVIDIA_RTX_PRO_6000_Blackwell"
RID="fix$(cat /proc/sys/kernel/random/uuid | cut -c1-6)"
OUT=/work/results; HOSTRES=~/bench/results
HF=$(grep -oE 'hf_[A-Za-z0-9]+' ~/.scherm_hf_token 2>/dev/null | head -1)
export HF_TOKEN="$HF"
log(){ echo "[$(date +%H:%M:%S)] $*"; }
dk(){ docker "$@"; }
free_port(){ local c; c=$(dk ps -q --filter "publish=8000"); [ -n "$c" ] && { dk rm -f $c >/dev/null 2>&1; sleep 3; }; }

serve(){
  local cname=$1 model=$2; shift 2
  free_port; dk rm -f "$cname" >/dev/null 2>&1
  dk run -d --name "$cname" --gpus all --ipc=host \
     -v ~/.cache/huggingface:/root/.cache/huggingface \
     -v "$HOSTRES":/work/results -v ~/bench/scripts:/scripts \
     -e HF_TOKEN -e PYTHONUNBUFFERED=1 -p 8000:8000 "$IMG" \
     vllm serve "$model" --served-model-name "$model" --port 8000 "$@" >/dev/null
  log "  aguardando $model (timeout 30min)..."
  for i in $(seq 1 180); do
    curl -sf -m3 http://localhost:8000/v1/models 2>/dev/null | grep -qF "\"$model\"" && { log "  ✅ $model PRONTO"; return 0; }
    dk ps -q -f name="$cname" | grep -q . || { log "  ❌ $cname morreu"; return 1; }
    sleep 10
  done
  log "  ❌ timeout $model"; return 1
}
chat(){
  local cname=$1 model=$2; shift 2
  log "########## serving $model (base-url SEM /v1) ##########"
  if serve "$cname" "$model" "$@"; then
    # base-url SEM /v1 — esse era o bug
    dk exec "$cname" python3 /scripts/run_serving.py --served-model "$model" --base-url http://localhost:8000 \
       --concurrencies 1 4 8 16 32 64 128 --reps 5 --input-len 512 --output-len 128 \
       --out "$OUT" --tag "$TAG" --run-id "$RID" 2>&1 | tail -8
  fi
  dk rm -f "$cname" >/dev/null 2>&1; sleep 4
}
chat c14  Qwen/Qwen2.5-14B-Instruct        --max-model-len 8192 --gpu-memory-utilization 0.85
chat ccod7 Qwen/Qwen2.5-Coder-7B-Instruct   --max-model-len 8192 --gpu-memory-utilization 0.85
chat c70  casperhansen/llama-3.3-70b-instruct-awq --max-model-len 8192 --gpu-memory-utilization 0.92 --quantization awq_marlin
log "########## SERVING FIX COMPLETO ##########"
echo "SERVING_FIX_COMPLETO"