providers/ollama.ps1
|
$POWERSHAI_OLLAMA = @{ ModelsCache = @{} } function Invoke-OllamaApi { <# .SYNOPSIS Base para invocar a API do ollama (parte da api que nao é compativel com a openai) #> [CmdletBinding()] param( $endpoint ,$body ,$method = 'POST' ,$token = $Env:OLLAMA_API_KEY ,$StreamCallback = $null ,[switch]$GetRawResp ) $Provider = Get-AiCurrentProvider $headers = @{} $OllamaBaseUrl = $POWERSHAI_SETTINGS.OllamaApiUrl; if($endpoint -match '^https?://'){ $url = $endpoint } else { verbose "Getting ApiURL" $url = GetCurrentProviderData ApiUrl $url += "/$endpoint" } verbose "OllamaBaseUrl: $url" $ReqParams = @{ data = $body url = $url method = $method Headers = $headers SseCallback = $StreamCallback } verbose "ReqParams:`n$($ReqParams|out-string)" try { $RawResp = InvokeHttp @ReqParams } catch [System.Net.WebException] { $ex = $_.exception; if($ex.PowershaiDetails){ $ResponseError = $ex.PowershaiDetails.ResponseError.text; if($ResponseError){ $err = New-PowershaiError "POWERSHAI_OLLAMA_ERROR" "Error: $ResponseError" -Prop @{ HttpResponseText = $ResponseError HttpResponse = $ex.PowershaiDetails.ResponseError.Response } throw $err } } throw; } verbose "RawResp: `n$($RawResp|out-string)" if($GetRawResp){ return $RawResp; } return $RawResp.text | ConvertFrom-Json } Set-Alias InvokeOllamaApi Invoke-OllamaApi # Change ollama urls! function Set-OllamaUrl { <# .SYNOPSIS Muda a URL do ollama. Formato protocolo://endereço:porta #> [CmdletBinding()] param( #Especifique a URL base somente (a url antes do /v1 ou /api). $url ) SetCurrentProviderData BaseUrl "$url/v1" # requerido pela openai SetCurrentProviderData ApiUrl "$url/api" } # Get all models! function Get-OllamaTags { <# .SYNOPSIS Lista as tags do ollama. Equivalente ao endpoint /api/tags .DESCRIPTION Mais info em https://github.com/ollama/ollama/blob/main/docs/api.md#list-local-models #> [CmdletBinding()] param() return (InvokeOllamaApi "tags" -m 'GET').models } Set-Alias Get-OllamaModels Get-OllamaTags Set-Alias OllamaTags Get-OllamaTags function Get-OllamaEmbeddings { <# .SYNOPSIS Obtém os embeddings usando um modelo de IA que suporta embeddings (Endpoint /api/embed) .DESCRIPTION Mais info: https://github.com/ollama/ollama/blob/main/docs/api.md#generate-embeddings #> param( $text ,# o modelo a ser usado # use Get-AiModels para uma lista de modelos que suportam embeddings! $model ) $DefaultModel = GetCurrentProviderData -Context DefaultEmbeddingsModel if(!$model){ $model = $DefaultModel } InvokeOllamaApi "embed" @{ model = $model input = $text } } function Get-OllamaModel { <# .SYNOPSIS Obtém detalhes de um modulo (/api/show) .DESCRIPTION Mais info em https://github.com/ollama/ollama/blob/main/docs/api.md#show-model-information #> [CmdletBinding()] param( # nome do modelo no formato modelo:tag (deve ser o nome exato) $model ) InvokeOllamaApi "show" -body @{ model = $model } } Set-Alias OllamaShow Get-OllamaModel Set-Alias Show-OllamaModel Get-OllamaModel function Update-OllamaModel { <# .SYNOPSIS Atualiza (ou baixa) um modelo do ollama (api/pull) .DESCRIPTION Mais info em https://github.com/ollama/ollama/blob/main/docs/api.md#pull-a-model #> [CmdletBinding()] param( # nome do modelo no formato modelo:tag (deve ser o nome exato) $model ,[switch]$Print ) $Stats = @{ StartTime = (Get-Date) StartTs = $null StartBytes = $null status = @() } $Sse = { param($data) if($Print){ write-host $data.line } try { $Event = $data.line | ConvertFrom-Json } catch { write-warning "Converting event status to json failed: $_"; write-host $Event; return; } if(!$Event.status){ return; } $PercentCompleted = 0; $CompletedBytes = $Event.completed; if(!$CompletedBytes){ $CompletedBytes = 0; } if($CompletedBytes){ [int]$PercentCompleted = ($CompletedBytes/$Event.Total)*100 } $PercentText = $PercentCompleted $TotalHuman = Bytes2Human $Event.total $CompletedHuman = Bytes2Human $Event.completed $HumanSpeed = $null; $EstimatedSeconds = $null if($Stats.StartBytes -eq $null){ $Stats.StartBytes = $Event.completed $Stats.StartTs = (Get-Date) } else { $Elapsed = (Get-Date) - $Stats.StartTs $CompletedDelta = $CompletedBytes - $Stats.StartBytes; $Speed = $CompletedDelta/$Elapsed.TotalSeconds; $HumanSpeed = Bytes2Human $Speed $LeftBytes = $Event.total - $Event.completed; if($Speed -gt 0){ $EstimatedSeconds = $LeftBytes/$Speed } else { $EstimatedSeconds = 0 $Speed = 0 $HumanSpeed = 0 } } if($Print){ write-host "`t" $Elapsed $CompletedDelta } $LastStatus = $Stats.status[-1] if($LastStatus -ne $Event.status){ $Stats.status += $Event.status } $EventId = $Stats.status.count; $ProgressParams = @{ Activity = $Event.status ID = $EventId } if($Event.completed){ $ProgressParams.PercentComplete = $PercentCompleted if($HumanSpeed){ $Status += "$HumanSpeed/s" } $Status = "$PercentText%","$CompletedHuman/$TotalHuman" if($EstimatedSeconds){ $ProgressParams.SecondsRemaining = $EstimatedSeconds } $ProgressParams.Status = $Status -Join " | " } write-progress @ProgressParams $Stats.LastRec = Get-Date; $Stats.LastCompleted = $CompletedBytes; } $resp = Invoke-OllamaApi 'pull' -method POST -body @{ model = $model } -StreamCallback $Sse -GetRawResp Reset-OllamaPowershaiCache } function ollama_GetEmbeddings { param( $text ,$model ,[switch]$IncludeText ,$dimensions ) $text = @($text); if($dimensions){ throw "POWERSHAI_OLLAMA_DIMENSIONS_NOTSUPPORTED: Ollama dont support specify dimensions." } verbose "Invoking native get embeddings..." $resp = Get-OllamaEmbeddings $text -model $model $i = -1; $resp.embeddings | %{ $i++; $emb = @{ embeddings = $_ } if($IncludeText){ $emb.text = $text[$i]; } return [PsCustomObject]$emb } } function ollama_FormatPrompt { param($model) $ModelEmoji = ""; if($model -like "llama*"){ $ModelEmoji = "🦙" } if($model -like "smollm*"){ $ModelEmoji = "🤗" } if($model -like "gemma*"){ $ModelEmoji = "💎" } if($model -like "aya*"){ $ModelEmoji = "❄️" } if($model -like "phi*"){ $ModelEmoji = "🟦" } if($model -like "qwen*"){ $ModelEmoji = "🟣" } return "⚪$($ModelEmoji) $($model): "; } function ollama_GetModels { $AllModels = Get-OllamaTags $Models = @(); foreach($model in $AllModels){ $ModelDetails = $POWERSHAI_OLLAMA.ModelsCache[$model.name] if(!$ModelDetails){ $ModelDetails = Get-OllamaModel $model.name $POWERSHAI_OLLAMA.ModelsCache[$model.name] = $ModelDetails } # Check tools based on that: https://github.com/ollama/ollama/blob/cdf3a181dcdb42ba72d9162c4f3461f218c33d5f/server/images.go#L98 # using that doc as variable source:https://github.com/ollama/ollama/blob/main/docs/template.md $ModelTemplate = $ModelDetails.template $SupportTools = $ModelTemplate -match '\.\b(Tools)\b' $ModelInfo = [PsCustomObject]@{ name = $model.name tools = $SupportTools tags = $model details = $ModelDetails } $Models += $ModelInfo; } return $Models; } function Reset-OllamaPowershaiCache { <# .SYNOPSIS Reseta o cache de modelos do ollama no powershai .DESCRIPTION O ollama mantém um cache de modelos para evitar consultar informações detalhadas a todo momento. Este comando reseta esse cache. Ao obter as informações da próxima vez (usando Get-AiModels), o cache é populado novamente. Este cache também é resetado automaticamente sempreque o powershai é importado novamente. #> [CmdletBinding()] param() $POWERSHAI_OLLAMA.ModelsCache = @{} } Set-Alias ollama_Chat Get-OpenAiChat # usa o mesmo da openai! function OllamaReqHooks { param($p) $MyData = $p.UserData; if($p.hook -eq "end"){ $c0 = $p.data.StreamData.FullAnswer.choices; if($c0){ $c0 = $c0[0] } else { return; } if($c0.delta.tool_calls){ $c0.finish_reason = "tool_calls" } } } return @{ RequireToken = $false BaseUrl = "http://localhost:11434/v1" ApiUrl = "http://localhost:11434/api" DefaultModel = $null CredentialEnvName = "OLLAMA_API_KEY" DefaultEmbeddingsModel = "nomic-embed-text:latest" EmbeddingsModels = @( "nomic-embed-.+" ) ReqHooks = (Get-Command OllamaReqHooks) info = @{ desc = "Permite acesso aos modelos disponibilizados pelo Ollama" url = "https://github.com/ollama/ollama" } IsOpenaiCompatible = $true } |