[mchinen]

The audio side is even more interesting, as it seems they totally got rid of positional embedding are just doing a single linear transform to match the LLM input dimension and that's it.

> Audio: We simplified audio processing even further. We removed the audio encoder entirely and projected the raw audio signal into the same dimensional space as text tokens.

[make3]

I guarantee you there's positional information one way or another. they just don't mention it because positional embeddings are extremely cheap computationally, not worth mentioning

[neosat]

Agree. Audio has strongly temporal so there is almost certainly some positional encoding one way or another.

[aesthesia]

Audio is 1 dimensional so the usual RoPE position encoding should handle it like it does for text tokens. You only need extra position encoding for higher-dimensional stuff like images.

[mchinen]

Ah yeah, thinking further it's probably just using some positioning embedding based on sequence numbering added in the LLM layers. For vision it needs the patch location as well.

[pseudollm]

No there isn't - read the paper. It's just 40msec raw audio samples. Multiplied by one matrix to translate to 3800 input vector. That's it. The next 40 msec are fed in the next transformer input step. Without any positional encoding. Repeat ad infinitum