disagg_dp_attn.sh 1.3 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
#!/bin/bash
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

# Setup cleanup trap
cleanup() {
    echo "Cleaning up background processes..."
    kill $DYNAMO_PID $PREFILL_PID 2>/dev/null || true
    wait $DYNAMO_PID $PREFILL_PID 2>/dev/null || true
    echo "Cleanup complete."
}
trap cleanup EXIT INT TERM

# run clear_namespace
15
python3 -m dynamo.sglang.utils.clear_namespace --namespace dynamo
16
17

# run ingress
18
python3 -m dynamo.frontend --http-port=8000 &
19
20
21
DYNAMO_PID=$!

# run prefill worker
22
python3 -m dynamo.sglang.worker \
23
24
25
26
27
28
29
30
31
32
33
34
35
  --model-path silence09/DeepSeek-R1-Small-2layers \
  --served-model-name silence09/DeepSeek-R1-Small-2layers \
  --tp 2 \
  --dp-size 2 \
  --enable-dp-attention \
  --trust-remote-code \
  --skip-tokenizer-init \
  --disaggregation-mode prefill \
  --disaggregation-transfer-backend nixl \
  --port 30000 &
PREFILL_PID=$!

# run decode worker
36
CUDA_VISIBLE_DEVICES=2,3 python3 -m dynamo.sglang.decode_worker \
37
38
39
40
41
42
43
44
45
  --model-path silence09/DeepSeek-R1-Small-2layers \
  --served-model-name silence09/DeepSeek-R1-Small-2layers \
  --tp 2 \
  --dp-size 2 \
  --enable-dp-attention \
  --trust-remote-code \
  --skip-tokenizer-init \
  --disaggregation-mode decode \
  --disaggregation-transfer-backend nixl \
46
  --port 31000