7 ms·
> New Causal Encoder–Decoder architecture: just 8B active parameters for input, 16B for output. Oh interesting, I can assume what the benefits is for including
by Alifatisk 6d ago
> New Causal Encoder–Decoder architecture: just 8B active parameters for input, 16B for output.
Oh interesting, I can assume what the benefits is for including the Encoder, but whats the downside? I’m thinking GPT (which is decoder only) ruled out Encoder for a reason?
- Alpha3031 6d agoEnc-decs are usually harder to train at frontier scale. Not 100% sure what DeepSeek has done differently here initial read seems to be something related to layer reuse but I just skimmed things so far.
- abecode 6d agoyes, that was surprising to me too. It would be a big deal if they switched to an encoder-decoder model like the original transformer. But I don't think that's what it's doing. One thing is the causal part, so in the original transformer, the encoder was bidirectional, but in this case it is not, so that's one difference. So I think it's an optimization for the prompt/prefill so that the attention is summarized into the output of the encoding layers, rather than all the layers. I just skimmed the paper too so if anyone else has insight, please correct me.