nova-loop

byKnk Ssmk

#!/usr/bin/env fish # corpus_loop.fish — Loop volunteer-compute with safety protocols # ── Config ──────────────────────────────────────────────────────────────────── set -l COOLDOWN_SEC 30 # Wait between runs (seconds) set -l TEMP_LIMIT 85 # GPU edge temp limit (°C) — pause if exceeded set -l LOG_FILE "$HOME/logofcro" set -gx HSA_OVERRIDE_GFX_VERSION 10.3.0 set -gx HIP_VISIBLE_DEVICES 0 # ── Setup ───────────────────────────────────────────────────────────────────── set -l total_runs 0 set -l session_start (date "+%Y-%m-%d %H:%M:%S") function log set -l msg "["(date "+%H:%M:%S")"] $argv" echo $msg echo $msg >> "$HOME/logofcro" end function check_gpu_temp set -l temp (rocm-smi --showtemp 2>/dev/null | grep "^0" | awk '{print $6}' | cut -d. -f1) if test -z "$temp" return 0 # Can't read temp, assume OK end if test $temp -ge $TEMP_LIMIT log "⚠️ GPU temp $temp°C exceeds limit ($TEMP_LIMIT°C) — cooling down 60s..." sleep 60 return 1 end return 0 end function on_interrupt log "🛑 Interrupted by user. Total runs: $total_runs" log "── Session ended ──────────────────────────────" exit 0 end trap on_interrupt INT # ── Main Loop ───────────────────────────────────────────────────────────────── log "══ corpus-client loop started (session: $session_start) ══" log " Cooldown: $COOLDOWN_SEC s | Temp limit: $TEMP_LIMIT°C" while true set total_runs (math $total_runs + 1) log "── Run #$total_runs starting ──" # Temperature check before each run check_gpu_temp # Run corpus-client uv run corpus-client volunteer-compute --limit 100 set -l exit_code $status if test $exit_code -eq log "✓ Run #$total_runs completed successfully" else log "✗ Run #$total_runs failed (exit $exit_code) — continuing..." end log " Cooling down $COOLDOWN_SEC s before next run..." sleep $COOLDOWN_SEC end evulate it and make sure that it should cross70 to 80c temp of gpu if it cross give collldown time