| #!/usr/bin/env bash |
| |
| |
| |
| LOG=/root/gate_flash_main.log |
| OUT=/root/autodl-tmp/graph-as-policy/outputs/gate_flash |
| log(){ echo "[$(date "+%F %T")] $*" >> "$LOG"; } |
| probe(){ K=$(cat /root/.gemini_key); c=$(curl -s -o /dev/null -w "%{http_code}" --max-time 20 -x http://127.0.0.1:18899 "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions" -H "Authorization: Bearer $K" -H "Content-Type: application/json" -d "{\"model\":\"gemini-3.1-flash-lite-preview\",\"messages\":[{\"role\":\"user\",\"content\":\"hi\"}],\"max_tokens\":3}"); [ "$c" = "200" ]; } |
| cd /root/autodl-tmp/graph-as-policy || exit 1 |
| mkdir -p "$OUT" |
| log "==== SELF-PACING GATE DRIVER START (waits for daily quota reset) ====" |
| for t in 0 1 2 3 4 5 6 7 8 9; do |
| [ -f "$OUT/t$t.done" ] && { log "task $t already clean -> skip"; continue; } |
| att=0 |
| while [ $att -lt 4 ]; do |
| att=$((att+1)) |
| while ! probe; do log "task $t att$att: quota/tunnel down, sleep 30m"; sleep 1800; done |
| log "task $t att$att: quota OK -> running" |
| source /root/repro_gemini_env.sh |
| rm -rf "$OUT/t$t" |
| uv run gap benchmark examples/benchmark/solo_flash_t${t}.yaml --output-dir "$OUT/t$t" > /root/gate_flash_t${t}.log 2>&1 |
| a=$(find "$OUT/t$t" -type d -name "trial_*reward*" 2>/dev/null | wc -l) |
| p=$(find "$OUT/t$t" -type d -name "trial_*reward1*pass" 2>/dev/null | wc -l) |
| q=$(grep -c "backend unavailable" /root/gate_flash_t${t}.log 2>/dev/null) |
| if [ "$a" -ge 3 ] && [ "$q" -eq 0 ]; then |
| touch "$OUT/t$t.done"; log "task $t CLEAN: pass=$p/$a (att$att)"; break |
| fi |
| log "task $t att$att incomplete (trials=$a quota_aborts=$q) -> retry"; sleep 300 |
| done |
| [ -f "$OUT/t$t.done" ] || log "task $t GAVE UP after 4 attempts" |
| sleep 120 |
| done |
| TP=0; TA=0 |
| for t in 0 1 2 3 4 5 6 7 8 9; do |
| a=$(find "$OUT/t$t" -type d -name "trial_*reward*" 2>/dev/null | wc -l) |
| p=$(find "$OUT/t$t" -type d -name "trial_*reward1*pass" 2>/dev/null | wc -l) |
| log " task $t: $p/$a"; TP=$((TP+p)); TA=$((TA+a)) |
| done |
| log "==== AGGREGATE: $TP/$TA passed ====" |
|
|
