#!/usr/bin/env bash
# 生成题推理 + 与 benchmark 基线对比
# 职责：对 240 道生成题跑推理，然后自动与 benchmark 基线做 Fisher 精确检验对比。
# 前提：benchmark 基线已跑完（workspaces/default/harness.db 中有 infer_adhoc）。
# 用法:
#   bash scripts/infer_generated.sh                                        # 全量 240 题 + 对比
#   N_SAMPLES=10 bash scripts/infer_generated.sh                           # smoke test
#   SKILLS_VERSION=v2 PROMPTS_VERSION=v2 bash scripts/infer_generated.sh   # 指定版本
#   BASELINE_RUN_ID=infer_v2 bash scripts/infer_generated.sh               # 指定基线 run_id
set -euo pipefail

cd "$(dirname "$0")/.."

CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0}"
export CUDA_VISIBLE_DEVICES

export HF_HUB_OFFLINE=1
export TRANSFORMERS_OFFLINE=1
export PYTHONUNBUFFERED=1

set -a
source .env
set +a

PYTHON="$(conda run -n Video-Tree-TRM which python)"

RUN_ID="${RUN_ID:-gen240}"
BASELINE_DB="${BASELINE_DB:-workspaces/default/harness.db}"
BASELINE_RUN_ID="${BASELINE_RUN_ID:-infer_adhoc}"

# ── Phase 1: 推理 ──
echo "========== Phase 1: 推理 240 道生成题 =========="

"${PYTHON}" main.py \
    --workspace-dir workspaces/default \
    --store-dir store \
    --mode infer \
    --concurrency 24 \
    --max-steps 40 \
    --skill-mode auto \
    --n-samples "${N_SAMPLES:-0}" \
    --questions generated \
    --run-id "${RUN_ID}" \
    --skills-version "${SKILLS_VERSION:-v1}" \
    --prompts-version "${PROMPTS_VERSION:-v1}"

# ── Phase 2: 对比 ──
echo "========== Phase 2: 与 benchmark 基线对比 =========="

"${PYTHON}" tools/generate_questions.py calibrate \
    --baseline-db "${BASELINE_DB}" \
    --baseline-run-id "${BASELINE_RUN_ID}" \
    --target-db workspaces/default/harness.db \
    --target-run-id "infer_${RUN_ID}"