dynamo/examples/backends/vllm/launch/agg_spec_decoding.sh

28 lines
903 B
Bash
Executable File

#!/bin/bash
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
set -e
trap 'echo Cleaning up...; kill 0' EXIT
# ---------------------------
# 1. Frontend (Ingress)
# ---------------------------
python -m dynamo.frontend --http-port=8000 &
# ---------------------------
# 2. Speculative Main Worker
# ---------------------------
# This runs the main model with EAGLE as the draft model for speculative decoding
DYN_SYSTEM_ENABLED=true DYN_SYSTEM_PORT=8081 \
CUDA_VISIBLE_DEVICES=0 python -m dynamo.vllm \
--model meta-llama/Meta-Llama-3.1-8B-Instruct \
--enforce-eager \
--speculative_config '{
"model": "yuhuili/EAGLE3-LLaMA3.1-Instruct-8B",
"draft_tensor_parallel_size": 1,
"num_speculative_tokens": 2,
"method": "eagle3"
}' \
--gpu-memory-utilization 0.8