start_api_server_rocm.sh

#!/usr/bin/env bash
# ACE-Step REST API Server Launcher - Linux AMD ROCm
# For AMD RX 7000/6000 series GPUs on Linux
# Requires: ROCm 6.x+ and ROCm PyTorch from https://pytorch.org/
#
# Setup:
#   1. Install ROCm: https://rocm.docs.amd.com/projects/install-on-linux/
#   2. Create venv:  python3 -m venv venv_rocm
#   3. Activate:     source venv_rocm/bin/activate
#   4. Install PyTorch for ROCm:
#      pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.3
#   5. Install dependencies:
#      pip install -r requirements-rocm-linux.txt
#   6. Run this script: ./start_api_server_rocm.sh

set -euo pipefail
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"

# ==================== ROCm Configuration ====================
# Force PyTorch LM backend (bypasses nano-vllm flash_attn dependency)
export ACESTEP_LM_BACKEND="pt"

# RDNA3 GPU architecture override (RX 7900 XT/XTX, RX 7800 XT, etc.)
# Change to 11.0.1 for gfx1101 (RX 7700 XT, RX 7800 XT)
# Change to 11.0.2 for gfx1102 (RX 7600)
export HSA_OVERRIDE_GFX_VERSION="${HSA_OVERRIDE_GFX_VERSION:-11.0.0}"

# MIOpen: use fast heuristic kernel selection instead of exhaustive benchmarking
# Without this, first-run VAE decode hangs for minutes on each conv layer
export MIOPEN_FIND_MODE="FAST"

# HuggingFace tokenizer parallelism
export TOKENIZERS_PARALLELISM="false"

# ==================== Server Configuration ====================
HOST="127.0.0.1"
# HOST="0.0.0.0"
PORT=8001

# ==================== Model Configuration ====================
# API key for authentication (optional)
API_KEY=""
# API_KEY="--api-key sk-your-secret-key"

# Download source: auto, huggingface, modelscope
DOWNLOAD_SOURCE=""
# DOWNLOAD_SOURCE="--download-source modelscope"

# LLM (Language Model) initialization settings
# By default, LLM is auto-enabled/disabled based on GPU VRAM:
#   - <=6GB VRAM: LLM disabled (DiT-only mode)
#   - >6GB VRAM: LLM enabled
# Values: auto (default), true (force enable), false (force disable)
export ACESTEP_INIT_LLM=auto
# export ACESTEP_INIT_LLM=true
# export ACESTEP_INIT_LLM=false

# LM model path (optional, only used when LLM is enabled)
# Available models: acestep-5Hz-lm-0.6B, acestep-5Hz-lm-1.7B, acestep-5Hz-lm-4B
LM_MODEL_PATH=""
# LM_MODEL_PATH="--lm-model-path acestep-5Hz-lm-0.6B"

# Update check on startup (set to "false" to disable)
CHECK_UPDATE="true"
# CHECK_UPDATE="false"

# Skip model loading at startup (models will be lazy-loaded on first request)
# Set to "true" to start server quickly without loading models
# export ACESTEP_NO_INIT=false
# export ACESTEP_NO_INIT=true

# ==================== Venv Configuration ====================
VENV_DIR="${SCRIPT_DIR}/venv_rocm"

# ==================== Startup Update Check ====================
_startup_update_check() {
    [[ "$CHECK_UPDATE" != "true" ]] && return 0
    command -v git &>/dev/null || return 0
    cd "$SCRIPT_DIR" || return 0
    git rev-parse --git-dir &>/dev/null 2>&1 || return 0

    local branch commit remote_commit
    branch="$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo "main")"
    commit="$(git rev-parse --short HEAD 2>/dev/null || echo "")"
    [[ -z "$commit" ]] && return 0

    echo "[Update] Checking for updates..."

    local fetch_ok=0
    if command -v timeout &>/dev/null; then
        timeout 10 git fetch origin --quiet 2>/dev/null && fetch_ok=1
    else
        git fetch origin --quiet 2>/dev/null && fetch_ok=1
    fi

    if [[ $fetch_ok -eq 0 ]]; then
        echo "[Update] Network unreachable, skipping."
        echo
        return 0
    fi

    remote_commit="$(git rev-parse --short "origin/$branch" 2>/dev/null || echo "")"

    if [[ -z "$remote_commit" || "$commit" == "$remote_commit" ]]; then
        echo "[Update] Already up to date ($commit)."
        echo
        return 0
    fi

    echo
    echo "========================================"
    echo "  Update available!"
    echo "========================================"
    echo "  Current: $commit  ->  Latest: $remote_commit"
    echo
    echo "  Recent changes:"
    git --no-pager log --oneline "HEAD..origin/$branch" 2>/dev/null | head -10
    echo

    read -rp "Update now before starting? (Y/N): " update_choice
    if [[ "${update_choice^^}" == "Y" ]]; then
        if [[ -f "$SCRIPT_DIR/check_update.sh" ]]; then
            bash "$SCRIPT_DIR/check_update.sh"
        else
            echo "Pulling latest changes..."
            git pull --ff-only origin "$branch" 2>/dev/null || {
                echo "[Update] Update failed. Please run: git pull"
            }
        fi
    else
        echo "[Update] Skipped. Run ./check_update.sh to update later."
    fi
    echo
}
_startup_update_check

echo "============================================"
echo "  ACE-Step 1.5 API - Linux AMD ROCm Edition"
echo "============================================"
echo

# Activate venv if it exists
if [[ -f "$VENV_DIR/bin/activate" ]]; then
    echo "Activating virtual environment: $VENV_DIR"
    source "$VENV_DIR/bin/activate"
else
    echo "WARNING: venv_rocm not found at $VENV_DIR"
    echo "Using system Python. See requirements-rocm-linux.txt for setup instructions."
fi
echo

# Verify ROCm PyTorch is installed
if ! python3 -c "
import torch
assert torch.cuda.is_available(), 'No GPU detected'
print(f'GPU: {torch.cuda.get_device_name(0)}')
hip = getattr(torch.version, 'hip', None)
print(f'HIP: {hip}' if hip else 'WARNING: Not a ROCm build')
" 2>/dev/null; then
    echo
    echo "========================================"
    echo " ERROR: ROCm PyTorch not detected!"
    echo "========================================"
    echo
    echo "Please install ROCm PyTorch first."
    echo "See requirements-rocm-linux.txt for instructions."
    echo
    echo "Quick setup:"
    echo "  python3 -m venv venv_rocm"
    echo "  source venv_rocm/bin/activate"
    echo "  pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.3"
    echo "  pip install -r requirements-rocm-linux.txt"
    echo
    exit 1
fi
echo

echo "Starting ACE-Step REST API Server..."
echo "API will be available at: http://${HOST}:${PORT}"
echo "API Documentation: http://${HOST}:${PORT}/docs"
echo

# Build command with optional parameters
CMD="--host $HOST --port $PORT"
[[ -n "$API_KEY" ]] && CMD="$CMD $API_KEY"
[[ -n "$DOWNLOAD_SOURCE" ]] && CMD="$CMD $DOWNLOAD_SOURCE"
[[ -n "$LM_MODEL_PATH" ]] && CMD="$CMD $LM_MODEL_PATH"

cd "$SCRIPT_DIR" && python3 -u acestep/api_server.py $CMD