50 lines
1.8 KiB
PowerShell
50 lines
1.8 KiB
PowerShell
[CmdletBinding()]
|
|
param(
|
|
[string]$OllamaUrl = 'http://127.0.0.1:11434',
|
|
[string]$OutputPath = ''
|
|
)
|
|
|
|
$ErrorActionPreference = 'Stop'
|
|
$endpoint = $OllamaUrl.TrimEnd('/') + '/api/ps'
|
|
$runtime = Invoke-RestMethod -Uri $endpoint -TimeoutSec 15
|
|
$models = @($runtime.models | ForEach-Object {
|
|
[ordered]@{
|
|
name = $_.name
|
|
model = $_.model
|
|
loaded_size_bytes = $_.size
|
|
vram_size_bytes = $_.size_vram
|
|
parameter_size = $_.details.parameter_size
|
|
quantization = $_.details.quantization_level
|
|
context_length = $_.context_length
|
|
expires_at = $_.expires_at
|
|
}
|
|
})
|
|
$runners = @(Get-Process -Name 'llama-server' -ErrorAction SilentlyContinue | ForEach-Object {
|
|
[ordered]@{
|
|
process_id = $_.Id
|
|
working_set_bytes = $_.WorkingSet64
|
|
private_bytes = $_.PrivateMemorySize64
|
|
}
|
|
})
|
|
|
|
$report = [ordered]@{
|
|
collected_at_local = [DateTimeOffset]::Now.ToString('o')
|
|
source = $endpoint
|
|
loaded_models = $models
|
|
llama_server_processes = $runners
|
|
note = 'Point-in-time snapshot, not peak memory. Process-to-model mapping and total host RAM are not asserted.'
|
|
}
|
|
|
|
if (-not $OutputPath) {
|
|
$resultsDirectory = Join-Path $PSScriptRoot '..\evals\results'
|
|
$stamp = Get-Date -Format 'yyyy-MM-dd_HHmmss'
|
|
$OutputPath = Join-Path $resultsDirectory "ollama_runtime_$stamp.json"
|
|
}
|
|
|
|
$resolvedOutput = [System.IO.Path]::GetFullPath($OutputPath)
|
|
$outputDirectory = Split-Path -Parent $resolvedOutput
|
|
New-Item -ItemType Directory -Force -Path $outputDirectory | Out-Null
|
|
$report | ConvertTo-Json -Depth 8 | Set-Content -LiteralPath $resolvedOutput -Encoding utf8
|
|
Write-Output "Saved runtime snapshot: $resolvedOutput"
|
|
Write-Output ("Loaded models: {0}; llama-server processes: {1}" -f $models.Count, $runners.Count)
|