Private/Get-OptimalMaxTokens.ps1
|
function Get-OptimalMaxTokens { <# .SYNOPSIS Calculates optimal MaxTokens based on prompt size and desired response size. .DESCRIPTION Analyzes the prompt and returns recommended MaxTokens value based on: - Estimated input token count - Desired response size (Tiny/Small/Medium/Large/Huge/Maximum) - Model context window limits .PARAMETER Prompt The prompt text to analyze .PARAMETER ResponseSize Desired response size. Default is 'Medium'. - Tiny: 300 tokens (~225 words, ~1 paragraph) - Small: 1000 tokens (~750 words, ~1 page) - Medium: 2000 tokens (~1,500 words, ~3 pages) - Large: 4096 tokens (~3,000 words, ~6 pages) - Huge: 8000 tokens (~6,000 words, ~12 pages) - Maximum: 16000 tokens (~12,000 words, ~24 pages) .PARAMETER ModelContextWindow Total context window of the model. Default is 128000 (for GPT-4o/Turbo). .EXAMPLE Get-OptimalMaxTokens -Prompt "What is Docker?" Returns: 2000 (Medium default) .EXAMPLE Get-OptimalMaxTokens -Prompt $largePrompt -ResponseSize "Huge" Returns: 8000 .EXAMPLE $maxTokens = Get-OptimalMaxTokens -Prompt $prompt -ResponseSize "Large" Use calculated value in API call #> [CmdletBinding()] [OutputType([int])] param( [Parameter(Mandatory)] [string]$Prompt, [Parameter()] [ValidateSet('Tiny', 'Small', 'Medium', 'Large', 'Huge', 'Maximum')] [string]$ResponseSize = 'Medium', [Parameter()] [int]$ModelContextWindow = 128000 ) # Estimate input tokens (1 token ≈ 4 characters for English) $estimatedInputTokens = [Math]::Ceiling($Prompt.Length / 4) # Response size presets $responseSizeMap = @{ 'Tiny' = 300 'Small' = 1000 'Medium' = 2000 'Large' = 4096 'Huge' = 8000 'Maximum' = 16000 } $desiredMaxTokens = $responseSizeMap[$ResponseSize] # Ensure we don't exceed model context window # Reserve 500 tokens as safety buffer $availableTokens = $ModelContextWindow - $estimatedInputTokens - 500 if ($availableTokens -le 0) { return $ModelContextWindow } # Use the smaller of desired tokens or available tokens $optimalMaxTokens = [Math]::Min($desiredMaxTokens, $availableTokens) Write-Verbose "Input tokens (estimated): $estimatedInputTokens" Write-Verbose "Desired MaxTokens: $desiredMaxTokens" Write-Verbose "Available tokens: $availableTokens" Write-Verbose "Optimal MaxTokens: $optimalMaxTokens" return $optimalMaxTokens } $script:headers = @{ "Content-Type" = "application/json" "X-Username" = "$($env:username)-$($env:userdnsdomain)" "X-Machine" = "$($env:computername)-$($env:userdnsdomain)" } |