DeepSeek-V3.2-Exp · Causal mask

61 transformer blocks › 58 MoE blocks

Causal mask

supplied · no parametersmask

Current and earlier key positions for every query; the indexer refines it.

shape: $M$ [T queries, T keys]

Sources

Contextual evidence from Construct causal mask

The model embeds tokens, applies all blocks with a causal mask, normalizes and projects to vocabulary logits with an untied head. The reference generator projects only the last position; training projects every position.

transformer · deepseek-ai/DeepSeek-V3.2-Exp / inference/model.py
Pinned revision 87e509a2e5a1 ↗ · lines 880–908

Read code excerpt

        self.embed = ParallelEmbedding(args.vocab_size, args.dim)
        self.layers = torch.nn.ModuleList()
        for layer_id in range(args.n_layers):
            self.layers.append(Block(layer_id, args))
        self.norm = RMSNorm(args.dim)
        # lm_head in the checkpoint is stored in bf16, while the parameter here is stored in fp32 for easier computation of logits later.
        self.head = ColumnParallelLinear(args.dim, args.vocab_size, dtype=torch.float32)
        self.register_buffer("freqs_cis", precompute_freqs_cis(args), persistent=False)

… lines 888–900 omitted …

        seqlen = tokens.size(1)
        freqs_cis = self.freqs_cis[start_pos:start_pos+seqlen]
        mask = torch.full((seqlen, seqlen), float("-inf"), device=tokens.device).triu_(1) if seqlen > 1 else None
        h, residual = self.embed(tokens), None
        for layer in self.layers:
            h, residual = layer(h, residual, start_pos, freqs_cis, mask)
        h, _ = self.norm(h, residual)
        logits = self.head(h[:, -1].float())

Contextual evidence from Multi-head latent attention

MLA owns a query down-projection, query RMSNorm and up-projection, a joint key–value down-projection that also emits the shared rotary key, a latent RMSNorm, a key–value up-projection and an output projection.

mla-init · deepseek-ai/DeepSeek-V3.2-Exp / inference/model.py
Pinned revision 87e509a2e5a1 ↗ · lines 523–539

Read code excerpt

        self.qk_head_dim = args.qk_nope_head_dim + args.qk_rope_head_dim
        self.v_head_dim = args.v_head_dim

        self.wq_a = Linear(self.dim, self.q_lora_rank)
        self.q_norm = RMSNorm(self.q_lora_rank)
        self.wq_b = ColumnParallelLinear(self.q_lora_rank, self.n_heads * self.qk_head_dim)
        self.wkv_a = Linear(self.dim, self.kv_lora_rank + self.qk_rope_head_dim)
        self.kv_norm = RMSNorm(self.kv_lora_rank)
        self.wkv_b = ColumnParallelLinear(self.kv_lora_rank, self.n_heads * (self.qk_nope_head_dim + self.v_head_dim))
        self.wo = RowParallelLinear(self.n_heads * self.v_head_dim, self.dim)
        self.softmax_scale = self.qk_head_dim ** -0.5
        self.scale_fmt = args.scale_fmt
        if args.max_seq_len > args.original_seq_len:
            mscale = 0.1 * args.mscale * math.log(args.rope_factor) + 1.0
            self.softmax_scale = self.softmax_scale * mscale * mscale

        self.indexer = Indexer(args)

Contextual evidence from Multi-head latent attention

In the full-sequence (prefill) path, the latent is up-projected into per-head keys and values; each head’s key joins its non-rotated part with the shared rotary key; scores are scaled, masked by the indexer’s selection and the causal mask, normalized and used to mix values.

mla-prefill · deepseek-ai/DeepSeek-V3.2-Exp / inference/model.py
Pinned revision 87e509a2e5a1 ↗ · lines 574–589

Read code excerpt

        if mask is not None:    # MHA prefill
            q = torch.cat([q_nope, q_pe], dim=-1)
            kv = self.wkv_b(kv)
            kv = kv.view(bsz, seqlen, self.n_local_heads, self.qk_nope_head_dim + self.v_head_dim)
            k_nope, v = torch.split(kv, [self.qk_nope_head_dim, self.v_head_dim], dim=-1)
            k = torch.cat([k_nope, k_pe.expand(-1, -1, self.n_local_heads, -1)], dim=-1)
            scores = torch.einsum("bshd,bthd->bsht", q, k).mul_(self.softmax_scale)

            # indexer
            topk_indices = self.indexer(x, qr, start_pos, freqs_cis, mask)
            index_mask = torch.full((bsz, seqlen, seqlen), float("-inf"), device=x.device).scatter_(-1, topk_indices, 0)
            index_mask += mask
            scores += index_mask.unsqueeze(2)

            scores = scores.softmax(dim=-1)
            x = torch.einsum("bsht,bthd->bshd", scores, v)

Contextual evidence from Multi-head latent attention

Head outputs are flattened and projected back to the model width.

mla-output · deepseek-ai/DeepSeek-V3.2-Exp / inference/model.py
Pinned revision 87e509a2e5a1 ↗ · lines 607–608

Read code excerpt

        x = self.wo(x.flatten(2))
        return x