LLamaSharp.SemanticKernel Abnormal conversation

[ynanech]

Role conversion may fail due to capitalization reasons

var role = Enum.TryParse<global::LLama.Common.AuthorRole>(chat.Role.Label, out var _role) ? _role : global::LLama.Common.AuthorRole.Unknown;

Main problem: When I only send one message, it will result in multiple rounds of conversation, which I really don't understand here

project reference

<ItemGroup> <PackageReference Include="Microsoft.SemanticKernel" Version="1.0.0-beta5" /> <PackageReference Include="LLamaSharp" Version="0.7.0" /> <PackageReference Include="LLamaSharp.Backend.Cpu" Version="0.7.0" /> </ItemGroup>

[AsakusaRinne]

Seems to be related with #186 , @xbotter any idea about it? It's most likely the problem of anti-prompt but I'm not sure how semantic-kernel makes it work.

[xbotter]

If necessary system instructions are missing in the dialogue, the output format may indeed exceed the expected results. It seems that some preliminary prompt engineering needs to be done. I will take a closer look.

[ynanech]

Microsoft.SemanticKernel.Abstractions" Version="1.0.0-beta6
In the latest version of SemanticKernel, IChatStreamingResult and IChatResult changes have been made,LLamaSharp SemanticKernel library needs to be adjusted

[xbotter]

I noticed that the current LlamaSharpChatCompletion interface is using LlamaSharp's InteractiveExecutor model, which is not very suitable for stateless IChatCompletion. I will change it to stateless mode in the near future.

[AsakusaRinne]

This issue may have been solved by #341, could you please try it again with the master branch?

[martindevans]

Since it's been a while and this was probably fixed by #341 I'll close this issue now. If it's still a problem please don't hesitate to re-open it!