File size: 3,537 Bytes
778d47d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 | #!/bin/bash
# Phase F: Best-of-N + selector eval (production inference protocol).
# For each of (a)/(b)/(b')/(d) configs, run K=8 planner sampling + 1 validator + 1 fixer
# per planner candidate, then compute three metrics:
# - greedy: EX of the first (K=1) trajectory (matches earlier K=1 numbers)
# - pass@8: EX if ANY of the 8 candidates is correct (oracle upper bound)
# - selector-majority: EX of the SQL whose execution result is the most common non-empty result
# among executable candidates (rule-based selector, no extra model needed)
set -e
cd /home/datht/mats-sql-tist
EVAL_INPUT=data/sft_bird_with_evidence_dev_text2sql.json
[ ! -f "$EVAL_INPUT" ] && EVAL_INPUT=data/sft_bird_dev_with_evidence_text2sql_new.json
P_SFT=alignment-handbook/output/qwen-coder0.5b-bird-planner-collab-sft
V_SFT=alignment-handbook/output/qwen-coder0.5b-bird-validator-sft
F_SFT=alignment-handbook/output/qwen-coder0.5b-bird-fixer-sft
P_INDEP=alignment-handbook/output/qwen-coder0.5b-bird-planner-INDEP-orpo
P_COLLAB=alignment-handbook/output/qwen-coder0.5b-bird-planner-COLLAB-orpo
F_ORPO=alignment-handbook/output/qwen-coder0.5b-bird-fixer-orpo
VLLM=/home/datht/anaconda3/envs/llm/bin/vllm
run_eval_bestof8() {
local LABEL=$1 P_CKPT=$2 V_CKPT=$3 F_CKPT=$4
echo ""
echo "================================================="
echo "=== EVAL $LABEL (best-of-8 + selector) ==="
echo " planner=$P_CKPT"
echo " validator=$V_CKPT"
echo " fixer=$F_CKPT"
echo "================================================="
pkill -f "vllm serve" 2>/dev/null || true
sleep 6
CUDA_VISIBLE_DEVICES=0 $VLLM serve "$P_CKPT" \
--served-model-name planner --port 8100 --dtype bfloat16 \
--gpu-memory-utilization 0.45 --max-model-len 8192 \
> /tmp/collab_eval_p.log 2>&1 &
CUDA_VISIBLE_DEVICES=1 $VLLM serve "$V_CKPT" \
--served-model-name validator --port 8101 --dtype bfloat16 \
--gpu-memory-utilization 0.42 --max-model-len 8192 \
> /tmp/collab_eval_v.log 2>&1 &
CUDA_VISIBLE_DEVICES=1 $VLLM serve "$F_CKPT" \
--served-model-name fixer --port 8102 --dtype bfloat16 \
--gpu-memory-utilization 0.42 --max-model-len 8192 \
> /tmp/collab_eval_f.log 2>&1 &
for url in http://localhost:8100/v1/models http://localhost:8101/v1/models http://localhost:8102/v1/models; do
for i in {1..120}; do
curl --noproxy localhost -fs $url >/dev/null 2>&1 && break
sleep 5
done
done
OUT=eval_results/collab_BoN8_${LABEL}_bird_dev.jsonl
PYTHONPATH=. /home/datht/anaconda3/envs/mats/bin/python scripts/run_pipeline_rollouts.py \
--input_file "$EVAL_INPUT" \
--output_file "$OUT" \
--planner_host http://localhost:8100 \
--validator_host http://localhost:8101 \
--fixer_host http://localhost:8102 \
--K 8 --K_val 1 --K_fix 1 --temperature 0.7 --top_p 0.9 \
--max_questions -1 --n_threads 8
# Compute the three metrics from the rollout JSONL
/home/datht/anaconda3/envs/mats/bin/python scripts/compute_bestofn_metrics.py "$OUT" "$LABEL"
}
run_eval_bestof8 a_sft_only "$P_SFT" "$V_SFT" "$F_SFT"
run_eval_bestof8 b_planner_indep "$P_INDEP" "$V_SFT" "$F_SFT"
run_eval_bestof8 b_prime_planner_indep_with_fixer_orpo "$P_INDEP" "$V_SFT" "$F_ORPO"
run_eval_bestof8 d_planner_collab_with_fixer_orpo "$P_COLLAB" "$V_SFT" "$F_ORPO"
pkill -f "vllm serve" 2>/dev/null || true
echo "DONE_COLLAB_BESTOFN_EVAL"
|