Is xAI/Grok supported?

[MetacodeX7]

Hi,

I notice there's no provider library for xAI and Grok, but I was able to use OpenAI as a replacement until now.

Recently something must have changed on xAI's side, and I get two errors:

• In the request: Argument not supported: logit_bias - this happens because logit_bias is always sent as [ ] by default (see GenerateAsync on OpenAiChatModel), but xAI will accept the request only with no logit_bias or if it's null.
• In streamed response deltas: missing "finish_reason" during deserialization. finish_reason is always present but null on streaming deltas sent by OpenAI, whereas with xAI responses it's just omitted when null. Unfortunately, in CreateChatCompletionStreamResponseChoice, FinishReason is marked with JsonRequired, even if it's nullable.

In the first case, I think xAI should be more open minded and just ignore the extra logit_bias.
In the second case, I think it's fair that a finish_reason would be missing when the streaming is not finished.
In both cases the OpenAi provider could be modified just a little to handle these variations gracefully (OpenAI doesn't care if logit_bias is null)

So my question is: would it make sense to adjust the OpenAi provider, so it supports those minor variations and handles xAI Grok requests?

[HavenDV]

Hi, I think it primarily comes from https://github.com/tryAGI/OpenAI SDK, but it seems I made the necessary adjustments to make Groq work, need to double check

Please note that the project does not have resources for active development now, it is better to use Microsoft.Extensions.AI / SemanticKernel if your requirements allow you

[MetacodeX7]

I made the necessary adjustments to make Groq work

Groq and Grok are two different products

Please note that the project does not have resources for active development now, it is better to use Microsoft.Extensions.AI / SemanticKernel if your requirements allow you

Understood, thanks! Following your reply, I switched to Llm Tornado to handle my Llm calls.