DeepSeek-V4.1-Flash
DeepSeekDeepSeek-V4.1-Flash adopts a Causal Encoder-Decoder (CED) architecture: a 40-layer Transformer organized as a 20-layer causal encoder followed by a 20-layer decoder.
4 more from this model
the KV entries are not derived from their respective hidden states ... Instead, they are projected directly from the hidden state of the (L/2)-th layer
we adopt a Causal Encoder-Decoder (CED) architecture, in which decoder global KV is projected from the final encoder hidden states
New Causal Encoder–Decoder architecture: just 8B active parameters for input, 16B for output.
the first built on the company's Causal Encoder-Decoder (CED) architecture