61 transformer blocks › 58 MoE blocks › Sparse latent attention
learnedself.q_norm(self.wq_a(x))
Compress each position’s 7,168 features into a 1,536-feature query latent.
Why
Queries pass through a low-rank latent as well; the authors say this can reduce activation memory during training. DeepSeek-V3 §2.1.1
- in
- [B batch, T positions, 7168 d]
- out
- [B batch, T positions, 1536 q latent]
- parameters
- 11,011,584 · × 58 MoE blocks = 638,671,872
Parameters
11,011,584 parameters per instance, across its learned components
| Component | Calculation | Parameters |
|---|---|---|
| Query down-projection | 11,010,048 | |
| Query latent RMSNorm | 1,536 | |
| Per-instance total | 11,011,584 | |
Across 58 independent MoE blocks: parameters.
Derived from the pinned reference configuration and inference code. Checkpoint tensors have not been independently counted; FP8 block scales and the multi-token-prediction module are not counted.
Explanation
Two matrices, 7,168 × 1,536 and 1,536 × 24,576, hold 48,758,784 weights; one 7,168 × 24,576 matrix would hold 176,160,768.
Codemodel.py 523–539, 560–564
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) qr = self.q_norm(self.wq_a(x))
q = self.wq_b(qr)
q = q.view(bsz, seqlen, self.n_local_heads, self.qk_head_dim)
q_nope, q_pe = torch.split(q, [self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
q_pe = apply_rotary_emb(q_pe, freqs_cis)Sources2
Direct evidence
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)Direct evidence
Queries are produced through a low-rank latent, split into a non-rotated and a rotated part, and rotated.
mla-queries · deepseek-ai/DeepSeek-V3.2-Exp / inference/model.py
Pinned revision 87e509a2e5a1 ↗ · lines 560–564
Read code excerpt
qr = self.q_norm(self.wq_a(x))
q = self.wq_b(qr)
q = q.view(bsz, seqlen, self.n_local_heads, self.qk_head_dim)
q_nope, q_pe = torch.split(q, [self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
q_pe = apply_rotary_emb(q_pe, freqs_cis)