disagg_kvbm_2p2d.sh 2.49 KB
Newer Older
1
#!/bin/bash
2
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3
4
5
6
7
# SPDX-License-Identifier: Apache-2.0
set -e
trap 'echo Cleaning up...; kill 0' EXIT

# run ingress with KV router
8
9
# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
python -m dynamo.frontend --router-mode kv &
10
11
12

# run decode workers on GPU 0 and 1, without enabling KVBM
# NOTE: remove --enforce-eager for production use
13
CUDA_VISIBLE_DEVICES=0 python3 -m dynamo.vllm --model Qwen/Qwen3-0.6B --kv-transfer-config '{"kv_connector":"NixlConnector","kv_role":"kv_both"}' --enforce-eager --disaggregation-mode decode &
14
VLLM_NIXL_SIDE_CHANNEL_PORT=20097 \
15
CUDA_VISIBLE_DEVICES=1 python3 -m dynamo.vllm --model Qwen/Qwen3-0.6B --kv-transfer-config '{"kv_connector":"NixlConnector","kv_role":"kv_both"}' --enforce-eager --disaggregation-mode decode &
16
17
18
19

# run prefill workers on GPU 2 and 3 with KVBM enabled using 20GB of CPU cache
# NOTE: use different barrier id prefixes for each prefill worker to avoid conflicts
# NOTE: remove --enforce-eager for production use
20
VLLM_NIXL_SIDE_CHANNEL_PORT=20098 \
21
22
23
24
DYN_KVBM_CPU_CACHE_GB=20 \
CUDA_VISIBLE_DEVICES=2 \
  python3 -m dynamo.vllm \
    --model Qwen/Qwen3-0.6B \
25
    --disaggregation-mode prefill \
26
    --kv-transfer-config '{"kv_connector":"PdConnector","kv_role":"kv_both","kv_connector_extra_config":{"connectors":[{"kv_connector":"DynamoConnector","kv_connector_module_path":"kvbm.vllm_integration.connector","kv_role":"kv_both"},{"kv_connector":"NixlConnector","kv_role":"kv_both"}]},"kv_connector_module_path":"kvbm.vllm_integration.connector"}' \
27
28
    --enforce-eager \
    --kv-events-config '{"publisher":"zmq","topic":"kv-events","endpoint":"tcp://*:20082","enable_kv_cache_events":true}' &
29

30
VLLM_NIXL_SIDE_CHANNEL_PORT=20099 \
31
32
DYN_KVBM_LEADER_ZMQ_PUB_PORT=56003 \
DYN_KVBM_LEADER_ZMQ_ACK_PORT=56004 \
33
34
35
36
DYN_KVBM_CPU_CACHE_GB=20 \
CUDA_VISIBLE_DEVICES=3 \
  python3 -m dynamo.vllm \
    --model Qwen/Qwen3-0.6B \
37
    --disaggregation-mode prefill \
38
    --kv-transfer-config '{"kv_connector":"PdConnector","kv_role":"kv_both","kv_connector_extra_config":{"connectors":[{"kv_connector":"DynamoConnector","kv_connector_module_path":"kvbm.vllm_integration.connector","kv_role":"kv_both"},{"kv_connector":"NixlConnector","kv_role":"kv_both"}]},"kv_connector_module_path":"kvbm.vllm_integration.connector"}' \
39
40
    --enforce-eager \
    --kv-events-config '{"publisher":"zmq","topic":"kv-events","endpoint":"tcp://*:20083","enable_kv_cache_events":true}'