71 lines
2.6 KiB
PowerShell
71 lines
2.6 KiB
PowerShell
param(
|
|
[string]$Engine = "$env:APPDATA\XianrenStudio\engines\cpu\llama-server.exe",
|
|
[string]$Model = "$env:APPDATA\XianrenStudio\models\qwen2.5-0.5b-instruct-q4_k_m.gguf",
|
|
[int]$Port = 8088
|
|
)
|
|
|
|
$ErrorActionPreference = "Stop"
|
|
if (-not (Test-Path $Engine)) { throw "engine not found: $Engine" }
|
|
if (-not (Test-Path $Model)) { throw "model not found: $Model" }
|
|
|
|
$log = Join-Path $env:TEMP "llama-server-smoke.log"
|
|
$args = @(
|
|
"--model", "`"$Model`"",
|
|
"--host", "127.0.0.1",
|
|
"--port", "$Port",
|
|
"--ctx-size", "2048",
|
|
"-ngl", "0",
|
|
"--no-webui"
|
|
)
|
|
|
|
$p = Start-Process -FilePath $Engine -ArgumentList $args `
|
|
-RedirectStandardOutput $log -RedirectStandardError "$log.err" `
|
|
-WindowStyle Hidden -PassThru
|
|
|
|
try {
|
|
$ready = $false
|
|
for ($i = 0; $i -lt 240; $i++) {
|
|
Start-Sleep -Milliseconds 500
|
|
try {
|
|
$r = Invoke-RestMethod -Uri "http://127.0.0.1:$Port/health" -TimeoutSec 2
|
|
if ($r.status -eq "ok") { $ready = $true; break }
|
|
} catch {}
|
|
}
|
|
if (-not $ready) {
|
|
Get-Content $log -ErrorAction SilentlyContinue | Select-Object -Last 20
|
|
throw "engine did not become ready; log: $log"
|
|
}
|
|
Write-Host "engine ready (pid $($p.Id), port $Port)"
|
|
|
|
$body = @{
|
|
model = "qwen-test"
|
|
messages = @(@{ role = "user"; content = "Introduce yourself in one sentence." })
|
|
stream = $false
|
|
max_tokens = 128
|
|
} | ConvertTo-Json -Depth 6
|
|
$resp = Invoke-RestMethod -Uri "http://127.0.0.1:$Port/v1/chat/completions" `
|
|
-Method Post -ContentType "application/json; charset=utf-8" -Body $body -TimeoutSec 120
|
|
Write-Host "NON-STREAM REPLY: $($resp.choices[0].message.content)"
|
|
|
|
$body2 = @{
|
|
model = "qwen-test"
|
|
messages = @(@{ role = "user"; content = "Count from 1 to 5." })
|
|
stream = $true
|
|
max_tokens = 64
|
|
} | ConvertTo-Json -Depth 6
|
|
$sse = Invoke-RestMethod -Uri "http://127.0.0.1:$Port/v1/chat/completions" `
|
|
-Method Post -ContentType "application/json; charset=utf-8" -Body $body2 -TimeoutSec 120
|
|
$hasDone = $sse -match "\[DONE\]"
|
|
$dataLines = ($sse -split "`n" | Where-Object { $_ -match "^data:" }).Count
|
|
Write-Host "STREAM: done-marker=$hasDone data-lines=$dataLines"
|
|
|
|
if (-not $hasDone -or $dataLines -lt 2) {
|
|
throw "streaming check failed"
|
|
}
|
|
Write-Host "SMOKE TEST PASSED"
|
|
} finally {
|
|
try { Invoke-RestMethod -Method Post -Uri "http://127.0.0.1:$Port/shutdown" -TimeoutSec 5 | Out-Null } catch {}
|
|
Start-Sleep -Seconds 1
|
|
if (-not $p.HasExited) { Stop-Process -Id $p.Id -Force -ErrorAction SilentlyContinue }
|
|
}
|