Skip to content

Latest commit

 

History

History
500 lines (410 loc) · 21.7 KB

CHANGELOG.md

File metadata and controls

500 lines (410 loc) · 21.7 KB
Raw

Changelog

4.11.0

  • The model name parameters are now case sensitive.
    Previously, model names were converted to lowercase internally. This is optimal for the OpenAI, but some compatible API servers require uppercase model names.
  • Add -All, -Limit & -Order params for Get-OpenAIFile.

4.10.0

4.9.1

  • Add a support for audio input / output in Request-ChatCompletion.

    Request-ChatCompletion -Model gpt-4o-audio-preview `
  -Modalities text, audio `
  -Voice 'shimmer' `
  -InputAudio 'C:\hello.mp3' `
  -AudioOutFile 'C:\response.mp3'

  • Fix an issue with incorrect discontinuance date for the gpt4-32k model.

4.8.0

  • Add an experimental support for Realtime API.
    This enables real-time conversations with the AI assitstant. Please refer to this guide.
    Guide: How to use Realtime API
  • Update default api version of Azure OpenAI Service to 2024-10-01-preview
  • Outputs Run and RunStep error messages as warnings.
  • Fix an issue where some request URIs contains duplicate slashes.

4.7.0

  • Add -Store & -MetaData params for Request-ChatCompletion.
    For support Stored Completions that was annouced at OpenAI DevDay 2024.

4.6.0

  • Add -Images parameter for Request-Moderation.

4.5.0

  • Add omni-moderation-latest model to tab completions.

4.4.1

  • Refactor file upload handling for non-ASCII filenames.

4.4.0

  • Add o1-preview and o1-mini models to tab completions.
  • Add -MaxCompletionTokens parameter for Request-ChatCompletion.
    The -MaxTokens parameter is now deprecated.
  • gpt-3.5-turbo-0613 and gpt-3.5-turbo-16k-0613 is deprecated on 2024-09-13.
    You can still call these models, but it is removed from model name completion.

4.3.0

  • Add -Include param for Get-ThreadRunStep.
  • Add -RankerForFileSearch and -ScoreThresholdForFileSearch params for New-Assistant.
  • Add -PollIntervalSec param for Wait-* functions.
  • Update default api version of Azure OpenAI Service to 2024-07-01-preview
  • Fix misc bugs related to Azure OpenAI Service.

4.2.0

  • Add support for new chatgpt-4o-latest model.

4.1.1

  • Fixed issue that occurred when using json_schema in Windows PowerShell 5.1 (#30)

4.1.0

4.0.0

This is a major release that includes breaking changes.

  • REMOVE All Azure specific function is removed.
    You can use Azure OpenAI Service by -ApiType parameter.
  • Add gpt-4o-mini model to tab completions.

3.16.0

  • Add -ServiceTier parameter to Request-ChatCompletion.

3.15.1

  • Fix issues about Get/Add/Remove-OpenAIFile with Azure OpenAI Service.
  • Fix issue that the New-VectorStore with -FileId parameter.

3.15.0

  • Add -ParallelToolCalls parameter.
  • Add -ChunkingStrategy parameter.
  • Add -MaxNumberOfFileSearchResults parameter.

3.14.0

  • Fix issue on Windows PowerShell 5.1 with Azure OpenAI Service.
  • Add -WaitForRunComplete param to Add-ThreadMessage. (#15) (Thanks @potatoqualitee!)

3.13.0

3.12.0

  • Support for o200k_base tokenizer that is used in gpt-4o.
  • Add -AsBatch for embeddings and text completions.
  • Add incomplete status of Run.

3.11.0

  • Add gpt-4o model to tab completions.

3.10.0

  • Add -Images parameter to Add-ThreadMessage.
    This makes that the image input can be used in Assistants.
  • Fix issues about relative file path input.
  • Rename Register-OpenAIFile to Add-OpenAIFile.
    (Note: You can still use Register-OpenAIFile as an alias. However, in the future release, the alias will be removed.)

3.9.1

  • Fix issue that the Start-ThreadRun would throw an authentication error in some situations. (#11) (Thanks @potatoqualitee!)
  • Fix issue that the Start-ThreadRun with -Stream doesn't work on Windows PowerShell 5.1
  • Improve documents.

3.9.0

  • Add new Remove-ThreadMessage function.
  • Add "required" as an option for -ToolChoice
  • Improve documents.

3.8.0

  • Commands for Assistants now uses v2 version of API. (Still in beta)
  • Add new commands for Vector Store.
  • Fix minor issues

Guide: How to use File search with Assistants and Vector Store

3.7.0

  • Add new functions for Batch API.

    To use Batch with the new PSOpenAI cmdlets, please refer to this guide.
    Guide: How to use Batch

    • Start-Batch
    • Get-Batch
    • Wait-Batch
    • Stop-Batch
    • Get-BatchOutput

  • Register-OpenAIFile is used to make byte arrays uploadable without saving them to a file.

    $ByteArray = [System.Text.Encoding]::UTF8.GetBytes('some text data')
Register-OpenAIFile -Content $ByteArray -Name 'filename.txt' -Purpose assistants

  • All Azure versions of the function is deprecated.
    Instead, the -ApiType parameter added to the normal functions.
    You can call the Azure OpenAI Service by specifying -ApiType as Azure.

    $env:OPENAI_API_KEY = '<Put your api key here>'
$env:OPENAI_API_BASE  = 'https://<your-resource-name>.openai.azure.com/'
Request-ChatCompletion `
  -Message 'Hello.' `
  -Deployment 'gpt-4' `
  -ApiType Azure

3.6.1

  • Fix a bug that caused New-Assistant to fail without explicitly passing the model name.
  • Fix an issue that Stop-ThreadRun does not working.
  • Improved code readability using splatting (#6) (Thanks @potatoqualitee!)

3.6.0

  • Add new params to Start-ThreadRun
    • -MaxPromptTokens
    • -MaxCompletionTokens
    • -TruncationStrategyType
    • -TruncationStrategyLastMessages
    • -ToolChoice
    • -ToolChoiceFunctionName
  • Updates the list of model names for tab completions.

3.5.0

  • Add -AdditionalMessages parameter to Start-ThreadRun
  • Add -RunId parameter to Get-ThreadMessage

3.4.0

  • Add -Temperature parameter to Start-ThreadRun

3.3.1

  • Remove unintended files in the previous release.

3.3.0

  • Change the parameter type to [switch] for -UseCodeInterpreter & -UseRetrieval
  • Add -Stream parameter to Start-ThreadRun
  • Add -Format parameter to Start-ThreadRun

3.2.0

  • Start-ThreadRun now can be use without preparing threads in advance.

    example :

    $Assistant = New-Assistant -Model "gpt-3.5-turbo"
$Run = Start-ThreadRun -Assistant $Assistant -Message "Hello, what can you do for me?"
$Result = $Run | Receive-ThreadRun -Wait

3.1.0

  • Change max value of the -TopLogProbs parameter to 20.
  • Change default api version of Azure OpenAI Service to 2024-03-01-preview
  • Add new common parameters to almost all functions.
    • -AdditionalQuery
    • -AdditionalHeaders
    • -AdditionalBody
  • Fix minor issues.

3.0.0

This is a major release that includes breaking changes.

  • REMOVE parameter alias Engine from all functions.
  • REMOVE old encoding support from ConvertTo-Token and ConvertFrom-Token. Now these commands only support cl100k_base encoding. It because all models currently supported by OpenAI use this encoding.
  • Remove Request-AzureExtensionsChatCompletion. It was undocumented.
  • Fix a bug that a file to be saved in an unexpected location when only a file name is specified in the -OutFile parameter.
  • Add new functions for using the Assistants API on the Azure OpenAI Service.
  • Add wav and pcm to response format of Request-AudioSpeech.

2.10.0

  • Add new embedding model's support to ConvertTo-Token.
  • Remove -InstanceId parameter to Request-ChatCompletion, it because the param does not GA.

2.9.0

  • Add -InstanceId parameter to Request-ChatCompletion.

2.8.0

  • Add -TimestampGranularities parameter to Request-AudioTranscription.
  • Add new Request-AzureAudioSpeech function.
  • Change default api version of Azure OpenAI Service to 2024-02-15-preview

2.7.0

  • The retry logic supports with retry-after-ms and retry-after response headers.
  • Add new model names to tab completions.
  • Add -Dimensions param for Request-Embeddings.
  • The exceptions thrown from API requests now have unique type information that differs depending on the error.
    Also, the exception contains the complete response header and contents, not just the message. This allows for more detailed information about the error.
    Currently, PSOpenAI has these type of exceptions.
    • APIRequestException
    • BadRequestException
    • ContentFilteredException
    • UnauthorizedException
    • NotFoundException
    • RateLimitExceededException
    • QuotaLimitExceededException

2.6.2

  • Improved performance when executing a large number of requests in a short period of time.

2.6.1

  • Fixed an issue where a double request was incorrectly executed when a -Stream was specified.

2.6.0

  • Remove Request-TextEdit function due to the API endpoint being shut down by OpenAI.
  • The retry decision respects the value of the x-should-retry response header.
  • Updates the list of deprecation models.
  • Rebuild libraries using .NET 8

2.5.0

  • Add -AdditionalInstructions parameter to Start-ThreadRun.

2.4.0

  • Add -LogProbs and -TopLogProbs parameter to Request-ChatCompletion.
  • Add -Images and -ImageDetail parameter to Request-AzureChatCompletion.
    These parameters are currently acceptable only on the gpt-4-vision-preview model.

2.3.0

  • Request-AzureImageGeneration now supports DALL-E 3 model.
  • Change default api version of Azure OpenAI Service to 2023-12-01-preview
  • Fix minor issues.

2.2.0

  • Add Request-AudioSpeech function. It generates audio from the input text.

    PS C:\> Request-AudioSpeech -text 'Do something fun to play.' -OutFile 'C:\Output\text2speech.mp3' -Voice Alloy

  • Add new functions for using the OpenAI Assistants API.

    To use Assistants with the new PSOpenAI cmdlets, please refer to this guide.
    Guide: How to use Assistants

Warning

The Assistants API is still in Beta. Specifications, usage, and parameters are subject to change without announcement.

  • Assistants: Get-Assistant, New-Assistant, Remove-Assistant, Set-Assistant
  • Threads: Get-Thread, New-Thread, Remove-Thread, Set-Thread
  • Messages: Get-ThreadMessage, Add-ThreadMessage
  • Runs: Get-ThreadRun, Start-ThreadRun, Stop-ThreadRun, Wait-ThreadRun, Receive-ThreadRun
  • Steps: Get-ThreadRunStep
  • Files: Get-OpenAIFile, Register-OpenAIFile, Remove-OpenAIFile, Get-OpenAIFileContent
  • Change directory name from "Examples" to "Guides".
  • Suppress verbose message about the organization-Id not found.
  • Fix minor bugs.

2.1.0

Implements some updates that announced in OpenAI Dev Day 2023.
(New features such as Threads or Assistants does not implemented yet. We are working in progress.)

  • Add / Rename / Remove some parameters to Request-ChatComplention.
    • Add : -Images, -ImageDetail, -Tools, -ToolChoice, -Format, -Seed
    • Remove : -Functions, -FunctionCall, -MaxFunctionCallCount
    • Rename : -InvokeFunctionOnCallMode to -InvokeTools
  • Add new models to tab completions of Request-ChatComplention.
    • gpt-3.5-turbo-1106, gpt-4-1106-preview, gpt-4-vision-preview
  • Add -Model parameter to Request-ImageGeneration. You can specifiy the model name. (dall-e-2 or dall-e-3)
  • Add -Quality and -Style parameters to Request-ImageGeneration (for dall-e-3)

2.0.0

This is a major release that includes breaking changes.

  • Add -ApiBase parameter to specify the base URL of the API endpoint.
    This is useful for using the OpenAI compatible API such like FastChat or LM Studio in a private environment. 
    PS C:\> Request-ChatCompletion -Message 'Hello' -ApiBase 'https://localhost:8000/v1'
  • Get/New/Remove-AzureOpenAIDeployments functions are removed.
  • Remove -Token parameter from all functions. Use -ApiKey instead.
  • OPENAI_TOKEN environment variable is deprecated. Use OPENAI_API_KEY instead.
  • The default model for Request-TextCompletion is changed to gpt-3.5-turbo-instruct from text-davinci-003.
  • Add -Format parameter to Request-Embeddings function. This parameter can be used to specify the format of the returned embeddings.

1.15.2

  • Fix various issues about Function calling.

1.15.0

  • Add new whisper functions for Azure OpenAI Service
  • Remove the upper limit on the -MaxTokens parameter. (The maximum number of tokens depends on the model used)
  • Add support for Function calling method on Azure OpenAI Service.
  • Change default api version of Azure OpenAI Service to 2023-09-01-preview

1.14.3

  • Add a new gpt-3.5-turbo-instruct model to tab completions of Request-TextCompletion.

1.14.2

  • Update tab completions for models of Request-TextCompletion.
    • To avoid breaking change, the default model used remains legacy text-davinci-003. We plan to change it on the future release.
  • Update warning messages about the deprecation models that has been announced by OpenAI.
  • Remove warning messages about the expiration dates of these models. Because OpenAI has announced that the discontinued dates for these models will be postponed for the time being.
    • gpt-3.5-turbo-0301
    • gpt-4-0314
    • gpt-4-32k-0314

1.14.0

  • Add initial support for Function calling method of the OpenAI ChatCompletion API. See the guide.
  • Change the name of -RolePrompt to -SystemMessage (-RolePrompt can also continue to be used as an alias )
  • Add new ChatGPT models.
    • gpt-3.5-turbo-16k
    • gpt-3.5-turbo-0613
    • gpt-3.5-turbo-16k-0613
    • gpt-4-0613
    • gpt-4-32k-0613
  • Ready for old models discontinuation on 2023-09-13.
    • gpt-3.5-turbo-0301
    • gpt-4-0314
    • gpt-4-32k-0314

1.13.0

1.12.6

  • Update Azure API version to 2023-05-15 of latest stable.
  • Update docs and examples.

1.12.5

  • Fix an issue that the User-Agent was not set in stream requests.
  • Minor fixes/improvements.

1.12.4

  • Use HTTP/2 for API requests if the platform is supported.
  • Adjust retry intervals for API requests.

1.12.3

  • Improve error handling.
  • The code-davinci-edit-001 model is restored.
    It appears that the previous unavailability was a temporary problem rather than a permanent retirement.

1.12.2

  • ConvertTo-Token and ConvertFrom-Token performance has been significantly improved. (Up to 100 times faster)

1.12.1

  • Improve error handling.

1.12.0

  • Messages can now be entered directly from the pipeline. (Request-ChatGPT) 
    PS C:\> "Can you recommend some music?" | Request-ChatGPT | Select-Object -ExpandProperty Answer
I can suggest some popular and diverse genres and artists that listeners enjoy...
  • Request-Moderation now outputs a warning message when content policies are violated. To suppress this, specify -WarningAction Ignore. 
    PS C:\> Request-Moderation -Text "This is a harmful message" -WarningAction Ignore
  • Fix incorrect links in the command help.

1.11.0

1.10.0

  • Add -Organization parameter to specify the Organization ID used for API requests.
  • Improve output of verbose and debug messages.
  • Enable tab completion for model names.
  • Misc improvements.

1.9.2

  • Add new commands Get-CosineSimilarity for calculates cosine similarity between two vectors.
    Note: Added for convenience, not good implementation for performance or accuracy. For production use, it is recommended to use an external library such as Math.NET Numerics.

1.9.1

  • Improve behavior of ConvertTo-Token and ConvertFrom-Token when a jagged array is input via pipeline.

1.9.0

1.8.0

  • Add -Name option for Request-ChatGPT.
    This parameter can be used to specify the name of the messenger.
    e.g.) 
    PS C:\> (Request-ChatGPT -Message 'Do you know my name?' -Name 'Samuel' -Model 'gpt-4-0314' -Temperature 0).Answer
Yes, your name is Samuel.
  • Change the -Message parameter of Request-ChatGPT to optional and accepts input from pipeline by property name.
  • Multiple strings can be specified for -RolePrompt.
  • Add -AsArray option for ConvertFrom-Token.
  • Some minor changes.

1.7.0

1.6.0

  • Add a new command Request-Embeddings.
  • [IMPORTANT CHANGE]
    Change the environment variable name of the API auth key from OPENAI_TOKEN to OPENAI_API_KEY. And also change the parameter name from -Token to -ApiKey.
    This is because the word "Token" is often confused with a term used in the field of machine learning, and the official OpenAI reference uses this name.
    For backward compatibility, OPENAI_TOKEN and -Token will continue to work, but may be deprecated entirely in the future.

1.5.0

  • Add -MaxRetryCount option for all functions.
    Retries up to the maximum number of times specified if an API request fails with a 429 (Rate limit reached) or 5xx (Server side errors) error. The retry interval increases exponentially up to 128 seconds.
  • These obsolated functions are completely removed.
    • Request-CodeCompletion
    • Request-CodeEdit
  • Minor fixes.

1.4.0

  • Add a new command Enter-ChatGPT. You can communicate with ChatGTP interactively on the console.

1.3.0

  • Add -Stream option for Request-ChatGPT and Request-TextCompletion.
  • The AI model code-davinci-edit-001 has been outdated.
  • Fix wrong messages about expires date of AI models.

1.2.1

  • OpenAI has announced that the Codex API will be discontinued on 2023-03-23, the following functions may no longer work in the future. In the future release, these functions will be completely removed from the module.
    • Request-CodeCompletion
    • Request-CodeEdit
  • When a request specifying an AI model that has been or will be discontinued by OpenAI is executed, Warning messages will be output. (the request continues to be executed).
    AI models that will output warning messages at this time:
    • code-davinci-001
    • code-davinci-002
    • code-cushman-001
    • code-cushman-002

1.2.0

  • Add StopSequence parameter for Request-ChatGPT, Request-TextCompletion, Request-CodeCompletion.
    When specific words are output from the API, subsequent output is stopped.
    e.g.) 
    # This code generates only top 4 list.
Request-TextCompletion -Prompt 'List of top 10 most populous countries' -StopSequence '5.'

1.1.2

  • Fix an issue that the language code may not be set correctly in Request-AudioTranscription on macOS and Linux environment.
  • Fix minor issues on Windows PowerShell 5.1
  • Fix Flaky Tests.

1.1.0

  • Improve error handling.
  • Change the default value of the MaxTokens parameter of Request-TextCompletion to 2048. The previous default value of 16 was not useful for most use cases.
  • Add the Request-CodeCompletion function. This performs the same process as Request-TextCompletion, but is more suitable for generating program code since the AI model used by default is code-davinci-002.
  • Add Pester tests for all public functions.

1.0.0

  • Initial public release.