agg_router_approx.sh 1.06 KB
Newer Older
1
2
3
4
#!/bin/bash
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

5
6
7
8
9
10
set -e
trap 'echo Cleaning up...; kill 0' EXIT

SCRIPT_DIR="$(dirname "$(readlink -f "$0")")"
source "$SCRIPT_DIR/../../../common/launch_utils.sh"

11
12
13
14
15
16
17
18
19
20
21
22
23
# Environment variables with defaults
export DYNAMO_HOME=${DYNAMO_HOME:-"/workspace"}
export MODEL_PATH=${MODEL_PATH:-"Qwen/Qwen3-0.6B"}
export SERVED_MODEL_NAME=${SERVED_MODEL_NAME:-"Qwen/Qwen3-0.6B"}
export AGG_ENGINE_ARGS=${AGG_ENGINE_ARGS:-"$DYNAMO_HOME/examples/backends/trtllm/engine_configs/qwen3/agg.yaml"}

# run frontend with KV router in approximate mode (i.e. no KV events)
python3 -m dynamo.frontend --router-mode kv --no-kv-events &

# run worker (no event publishing needed - frontend handles routing with predictive approx kv mode)
python3 -m dynamo.trtllm \
  --model-path "$MODEL_PATH" \
  --served-model-name "$SERVED_MODEL_NAME" \
24
25
26
27
  --extra-engine-args "$AGG_ENGINE_ARGS" &

# Exit on first worker failure; kill 0 in the EXIT trap tears down the rest
wait_any_exit