| #!/usr/bin/env bash |
| set -euo pipefail |
|
|
| TASK_IDS_FILE="${TASK_IDS_FILE:-training/benchmarks/cybergym_smoke_tasks.txt}" |
| AGENT_EXAMPLES_DIR="${AGENT_EXAMPLES_DIR:-/workspace/cybergym-agent-examples}" |
| OPENHANDS_REPO="${OPENHANDS_REPO:-$AGENT_EXAMPLES_DIR/openhands/openhands-repo}" |
| CYBERGYM_DATA_DIR="${CYBERGYM_DATA_DIR:-/workspace/cybergym_data/data}" |
| CYBERGYM_SERVER_URL="${CYBERGYM_SERVER_URL:-http://127.0.0.1:8666}" |
| OPENAI_BASE_URL="${OPENAI_BASE_URL:-http://127.0.0.1:8000/v1}" |
| MODEL="${MODEL:-qwen36-base/thinking}" |
| OUT_DIR="${OUT_DIR:-/workspace/reports/cybergym/qwen36_27b_base/smoke_10}" |
| DIFFICULTY="${DIFFICULTY:-level1}" |
| MAX_ITER="${MAX_ITER:-100}" |
| TIMEOUT="${TIMEOUT:-1200}" |
| SILENT="${SILENT:-false}" |
|
|
| if [[ ! -f "$TASK_IDS_FILE" ]]; then |
| echo "Missing task id file: $TASK_IDS_FILE" >&2 |
| exit 2 |
| fi |
|
|
| if [[ ! -f "$AGENT_EXAMPLES_DIR/openhands/run.py" ]]; then |
| echo "Missing OpenHands CyberGym runner: $AGENT_EXAMPLES_DIR/openhands/run.py" >&2 |
| exit 3 |
| fi |
|
|
| mkdir -p "$OUT_DIR/logs" "$OUT_DIR/tmp" |
| export LLM_API_KEY="${LLM_API_KEY:-EMPTY}" |
|
|
| while IFS= read -r task_id; do |
| [[ -z "$task_id" || "$task_id" == \ |
|
|
| echo "Running CyberGym baseline task: $task_id" |
| python3 "$AGENT_EXAMPLES_DIR/openhands/run.py" \ |
| --model "$MODEL" \ |
| --base_url "$OPENAI_BASE_URL" \ |
| --log_dir "$OUT_DIR/logs" \ |
| --tmp_dir "$OUT_DIR/tmp" \ |
| --repo "$OPENHANDS_REPO" \ |
| --data_dir "$CYBERGYM_DATA_DIR" \ |
| --task_id "$task_id" \ |
| --server "$CYBERGYM_SERVER_URL" \ |
| --timeout "$TIMEOUT" \ |
| --max_iter "$MAX_ITER" \ |
| --silent "$SILENT" \ |
| --difficulty "$DIFFICULTY" |
| done < "$TASK_IDS_FILE" |
|
|