D:\App\llama-b9840-bin-win-cuda-12.4-x64>wmic cpu get name
Name
AMD Ryzen 7 6800H with Radeon Graphics
可以给我设计 bat 了吗，或者你最好可以设计一套所有设备都通用的 bat 脚本，实在不行就设计为我专用的吧
llama-server.exe -m models\Qwythos-9B-Claude-Mythos-5-1M-MTP-Q4_K_M.gguf -ngl 12 -c 2048 --host 127.0.0.1 --port 8080# ==============================
# LingDuAI Smart Launcher v3
# Auto GPU Safe Mode
# ==============================
cd $PSScriptRoot
Write-Host "===================================="
Write-Host " LingDuAI Smart Launcher v3"
Write-Host "===================================="
# 1. 获取显存
$gpu = nvidia-smi --query-gpu=memory.total,memory.used --format=csv,noheader,nounits
$total, $used = $gpu -split ", "
$free = [int]$total - [int]$used
Write-Host "GPU Total: $total MB"
Write-Host "GPU Used : $used MB"
Write-Host "GPU Free : $free MB"
# 2. 自动策略
$ngl = 4
$ctx = 1024
$enableVision = $false
if ($free -gt 3000) {
$ngl = 6
$enableVision = $true
} elseif ($free -gt 2000) {
$ngl = 4
$enableVision = $true
} else {
$ngl = 2
$enableVision = $false
}
Write-Host "Selected NGL: $ngl"
Write-Host "Vision Mode: $enableVision"
Write-Host "CTX: $ctx"
# 3. 自动找模型
$model = Get-ChildItem models\*.gguf |
Where-Object { $_.Name -notmatch "mmproj" -and $_.Name -notmatch "clip" } |
Where-Object { $_.Name -match "Q4_K_M|Q5_K_M" } |
Select-Object -First 1
if (-not $model) {
$model = Get-ChildItem models\*.gguf |
Where-Object { $_.Name -notmatch "mmproj" } |
Select-Object -First 1
}
Write-Host "Model: $($model.Name)"
# 4. mmproj 自动选择
$mmproj = $null
if ($enableVision) {
$mmproj = Get-ChildItem models\mmproj*.gguf | Select-Object -First 1
if ($mmproj) {
Write-Host "Vision Model: $($mmproj.Name)"
}
}
# 5. 启动参数
$args = @(
"-m", "models\$($model.Name)",
"-ngl", "$ngl",
"-c", "$ctx",
"--host", "127.0.0.1",
"--port", "8080",
"--cont-batching"
)
if ($mmproj) {
$args += "--mmproj"
$args += "models\$($mmproj.Name)"
}
# 6. 启动
Write-Host "Starting server..."
Start-Process -FilePath ".\llama-server.exe" -ArgumentList $args -NoNewWindow -Wait