All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Fix default path to save PBE file in TikTokenSharp.
- Fill recorded message content with empty string because messages in request parameter must have content field.
- Fix dependencies in
package.json.
- Fix request to add
HttpCompletionOption.ResponseHeadersReadto Relent streaming API.
- Add Streaming API sample.
- Change streaming API to use
IAsyncEnumerable. - Change to
UniTaskfromTaskin Relent API.
- Add 16k context turbo model option.
- Add function calling support.
- Add optional
nameproperty toMessage. - Add verbose log option.
- Update chat models at 2023-06-13.
- Add dependencies to
package.json.
- Simplify package structure.
- Update Relent version to 0.2.0.
- Add extensions of
IChatMemory.
- Improve interface of
IChatMemoryfor async operation. - Be
IChatMemoryimplementations thread-safe.
- Fix capacity of
FiniteQueueChatMemory.
- Improve accessibility of
Messaage.
- Add
TiktokenSharpthat is a tokenizer to calculate token length of text in local.
- Add option to specify instance of
HttpClient. - Add option to receive response as
Stream.
- Improve null check of response content.
- Add GPT-4 all models.
- Add memory management of chat completion API by
IChatMemory. - Add resilient error handling implementation of chat completion API by Relent.
- Add GPT-4 model enum.
- Change name and arguments of completion method.
- Fix deserialization error of
"usage"in response body.
- Add optional request parameters.
- Fix error handling of API response.
- Implement ChatGPT chat completion API bindings to C#.
- Implement sample component of ChatGPT API.