disagg.sh 2.9 KB
Newer Older
1
#!/bin/bash
2
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3
4
# 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
# Environment variables with defaults
12
export DYNAMO_HOME=${DYNAMO_HOME:-"/workspace"}
13
14
export MODEL_PATH=${MODEL_PATH:-"Qwen/Qwen3-0.6B"}
export SERVED_MODEL_NAME=${SERVED_MODEL_NAME:-"Qwen/Qwen3-0.6B"}
15
16
export PREFILL_ENGINE_ARGS=${PREFILL_ENGINE_ARGS:-"$DYNAMO_HOME/examples/backends/trtllm/engine_configs/qwen3/prefill.yaml"}
export DECODE_ENGINE_ARGS=${DECODE_ENGINE_ARGS:-"$DYNAMO_HOME/examples/backends/trtllm/engine_configs/qwen3/decode.yaml"}
17
18
export PREFILL_CUDA_VISIBLE_DEVICES=${PREFILL_CUDA_VISIBLE_DEVICES:-"0"}
export DECODE_CUDA_VISIBLE_DEVICES=${DECODE_CUDA_VISIBLE_DEVICES:-"1"}
19
20
21
export MODALITY=${MODALITY:-"text"}
# If you want to use multimodal, set MODALITY to "multimodal"
#export MODALITY=${MODALITY:-"multimodal"}
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
ENABLE_OTEL=false
while [[ $# -gt 0 ]]; do
    case $1 in
        --enable-otel)
            ENABLE_OTEL=true
            shift
            ;;
        -h|--help)
            echo "Usage: $0 [OPTIONS]"
            echo "Options:"
            echo "  --enable-otel        Enable OpenTelemetry tracing"
            echo "  -h, --help           Show this help message"
            echo ""
            exit 0
            ;;
        *)
            echo "Unknown option: $1"
            echo "Use --help for usage information"
            exit 1
            ;;
    esac
done

# Enable tracing if requested
TRACE_ARGS=()
if [ "$ENABLE_OTEL" = true ]; then
    export DYN_LOGGING_JSONL=true
    export OTEL_EXPORT_ENABLED=1
    export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=${OTEL_EXPORTER_OTLP_TRACES_ENDPOINT:-http://localhost:4317}
    TRACE_ARGS+=(--override-engine-args "{\"return_perf_metrics\": true, \"otlp_traces_endpoint\": \"${OTEL_EXPORTER_OTLP_TRACES_ENDPOINT}\" }")
fi
54

55
# run frontend
56
# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
57
OTEL_SERVICE_NAME=dynamo-frontend \
58
python3 -m dynamo.frontend &
59
60

# run prefill worker
61
OTEL_SERVICE_NAME=dynamo-worker-prefill CUDA_VISIBLE_DEVICES=$PREFILL_CUDA_VISIBLE_DEVICES python3 -m dynamo.trtllm \
62
63
64
  --model-path "$MODEL_PATH" \
  --served-model-name "$SERVED_MODEL_NAME" \
  --extra-engine-args  "$PREFILL_ENGINE_ARGS" \
65
  --modality "$MODALITY" \
66
67
  --disaggregation-mode prefill \
  "${TRACE_ARGS[@]}" &
68
69

# run decode worker
70
OTEL_SERVICE_NAME=dynamo-worker-decode CUDA_VISIBLE_DEVICES=$DECODE_CUDA_VISIBLE_DEVICES python3 -m dynamo.trtllm \
71
72
73
  --model-path "$MODEL_PATH" \
  --served-model-name "$SERVED_MODEL_NAME" \
  --extra-engine-args  "$DECODE_ENGINE_ARGS" \
74
  --modality "$MODALITY" \
75
  --disaggregation-mode decode \
76
77
78
79
  "${TRACE_ARGS[@]}" &

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