blob: b658bab14ceb22db438d112611cdaa24210da3a6 (
plain)
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
|
#!/bin/bash
#SBATCH --job-name=exp_rag_vector_p100
#SBATCH --account=bfqt-delta-gpu
#SBATCH --partition=gpuA100x4
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=16
#SBATCH --gres=gpu:nvidia_a100:2
#SBATCH --mem=128G
#SBATCH --time=12:00:00
#SBATCH --output=exp_rag_vector_p100_%j.out
#SBATCH --error=exp_rag_vector_p100_%j.err
# Full run: rag_vector method, profiles 100-150 (50 profiles × 15 sessions)
cd /projects/bfqt/users/yurenh2/ml-projects/personalization-user-model
source /u/yurenh2/miniforge3/etc/profile.d/conda.sh
conda activate eval
export HF_HOME=/projects/bfqt/users/yurenh2/hf_cache/huggingface
export PYTHONPATH="${PWD}/src:${PWD}/collaborativeagents:${PYTHONPATH}"
export NCCL_P2P_DISABLE=1
set -a
source .env
set +a
pip install --quiet openai python-dotenv json-repair
MODEL_8B="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/models/llama-3.1-8b-instruct"
pkill -f "vllm.entrypoints" 2>/dev/null || true
sleep 2
CUDA_VISIBLE_DEVICES=0 python -m vllm.entrypoints.openai.api_server \
--model $MODEL_8B --port 8003 --tensor-parallel-size 1 \
--gpu-memory-utilization 0.5 --max-model-len 8192 \
--dtype bfloat16 --disable-log-requests &
for i in $(seq 1 90); do
curl -s http://localhost:8003/health > /dev/null 2>&1 && break
sleep 2
done
echo "vLLM ready."
cd collaborativeagents/scripts
python run_experiments.py \
--methods rag_vector \
--datasets math-hard \
--n-profiles 200 \
--start-profile 100 \
--end-profile 150 \
--n-sessions 15 \
--max-turns 8 \
--use-vllm \
--use-openai-user \
--openai-user-model gpt-5-mini \
--reward-mode llm \
--vllm-agent-url http://localhost:8003/v1 \
--parallel-profiles 25 \
--profile-path ../data/complex_profiles_v2/profiles_200.jsonl \
--output-dir ../results/fullscale_15sess
echo "rag_vector p100-150 complete!"
pkill -f "vllm.entrypoints" 2>/dev/null || true
|
