Aaron Friel

@youkaichao How well does the KV cache handle a block size of 1, in terms of compute or memory overhead?

Would you support pinned caching in the same manner as the OpenAI chat completion API as Anthropic? I'm not sure how you would support this for text completion APIs but...

@llsj14 The first thing that comes to mind is that explicit APIs add a new dimension to metering and cost analysis, but also security. If it's efficient to check if...

@jalmeroth thanks for reporting this. If I have this right, you have an @pulumi/actions job like so: ``` - uses: pulumi/actions@v3 with: config-map: some-key: value: - some-value-a - some-value-b ```...

It would be great if @yzhang-gh you could identify what might cause this feedback loop with other extensions. The issues you refer to, and the two that link to this...

I did some cursory research into the libraries available for cross-process IPC in Rust, and, well, essentially all of them add some significant degree of complexity to the implementation or...

## Root cause The AI backend is throwing an error when generating the zip template. After extracting the program from the markdown of the conversation, the backend also parses the...

That leaves us with a few issues: ## Incorrect code generation We can prompt and train the model to try to prevent this statistically. I'll open an issue and make...

Static methods also return `T` instead of `GuestT`, so this workaround doesn't work: ```wit interface foo { resource bar { fallible-ctor: static func() -> result } } ``` I think...

As a user of guidance/AICI/other methods of constraining LLM output, disabling beam search can reduce quality of outputs. For the reason users describe above. We've noticed that across a wide...