Kevin Wang

Interesting, really weird that Safari decided to create their own payload format. This has some rather odd implications for interoperability, however. Suddenly the payloader needs to know about where the...

> Ok will give it further thought. > > One way that comes to mind is something like this: > > ```c > int32_t llama_decode_with_sampling( > struct llama_context * ctx,...

Thanks, glad you enjoyed it! Yes, I think building H264 from scratch is my next goal. It's definitely feasible to build something compact as a learning exercise, but I need...

Yeah this sounds like something broken with the onnxruntime-web API. Does it work at 08d78db9a4aafb89302b0a6d5e813a50672fc2a6? Just one before the onnxruntime-web version bump.

Hi @JGCoelho thanks for trying it out! I made a mistake in the instructions, could you try the `magic-copy.zip` file in the latest release? https://github.com/kevmo314/magic-copy/releases I've updated the README as...

Does this happen when you install the extension locally? I am able to work off of local files. I did see that it breaks for the release version, I think...

It seems that we have to use a separate manifest file for Firefox because it doesn't support `background.service_worker` in Manifest V3 yet :(

> So this add-on is only for chrome? At the moment, yes. You can try the Firefox release but I haven't been able to test it thoroughly.

I'm also interested in the status of this, curious what's necessary to support LoRA inference with speculative decoding?

Screenshot for the behavior?