blob: 35f444043e5ce9bf3d779640e44c19ae61e3abaa (
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
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
|
#!/bin/bash
# Test batch processing with 50 conversations (paper's configuration)
set -e
cd /projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents
source /u/yurenh2/miniforge3/etc/profile.d/conda.sh
conda activate eval
export HF_HOME=/projects/bfqt/users/yurenh2/hf_cache/huggingface
export PYTHONPATH="${PWD}:${PWD}/../src:${PYTHONPATH}"
MODEL_8B="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/models/llama-3.1-8b-instruct"
PORT_USER=8004
PORT_AGENT=8003
echo "============================================"
echo "Batch Processing Test (Paper Configuration)"
echo "Batch Size: 50 conversations"
echo "============================================"
date
nvidia-smi --query-gpu=index,name,memory.total --format=csv
echo ""
# Kill any existing vLLM servers
pkill -f "vllm.entrypoints" 2>/dev/null || true
sleep 2
# Start servers with TP=2
echo "Starting 8B user simulator server (GPU 0-1, TP=2)..."
CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
--model $MODEL_8B \
--port $PORT_USER \
--tensor-parallel-size 2 \
--gpu-memory-utilization 0.85 \
--max-model-len 4096 \
--disable-log-requests \
--dtype bfloat16 &
SERVER_USER_PID=$!
echo "Starting 8B agent server (GPU 2-3, TP=2)..."
CUDA_VISIBLE_DEVICES=2,3 python -m vllm.entrypoints.openai.api_server \
--model $MODEL_8B \
--port $PORT_AGENT \
--tensor-parallel-size 2 \
--gpu-memory-utilization 0.85 \
--max-model-len 4096 \
--disable-log-requests \
--dtype bfloat16 &
SERVER_AGENT_PID=$!
echo "Waiting for servers..."
for i in $(seq 1 100); do
READY_USER=$(curl -s http://localhost:$PORT_USER/health > /dev/null 2>&1 && echo 1 || echo 0)
READY_AGENT=$(curl -s http://localhost:$PORT_AGENT/health > /dev/null 2>&1 && echo 1 || echo 0)
if [ "$READY_USER" = "1" ] && [ "$READY_AGENT" = "1" ]; then
echo "Both servers ready after $((i*3)) seconds"
break
fi
if [ $((i % 20)) -eq 0 ]; then
echo " Still waiting... ($((i*3))s)"
fi
sleep 3
done
if ! curl -s http://localhost:$PORT_USER/health > /dev/null; then
echo "ERROR: User server not healthy"; kill $SERVER_USER_PID $SERVER_AGENT_PID 2>/dev/null; exit 1
fi
if ! curl -s http://localhost:$PORT_AGENT/health > /dev/null; then
echo "ERROR: Agent server not healthy"; kill $SERVER_USER_PID $SERVER_AGENT_PID 2>/dev/null; exit 1
fi
echo "✓ Both servers healthy"
echo ""
echo "============================================"
echo "Test 1: Batch=50, Turns=5"
echo "============================================"
python scripts/test_batch_50.py \
http://localhost:$PORT_USER/v1 \
http://localhost:$PORT_AGENT/v1 \
50 5
echo ""
echo "============================================"
echo "Test 2: Batch=50, Turns=10 (paper config)"
echo "============================================"
python scripts/test_batch_50.py \
http://localhost:$PORT_USER/v1 \
http://localhost:$PORT_AGENT/v1 \
50 10
echo ""
echo "============================================"
echo "Test 3: Batch=100, Turns=10 (stress test)"
echo "============================================"
python scripts/test_batch_50.py \
http://localhost:$PORT_USER/v1 \
http://localhost:$PORT_AGENT/v1 \
100 10
# Cleanup
echo ""
echo "Cleaning up..."
kill $SERVER_USER_PID $SERVER_AGENT_PID 2>/dev/null || true
echo ""
date
|
