# LLaMA-3 Block vs DeepSeek-V3

What a frontier open model adds to the block everyone started from.

**DeepSeek-V3 has 666B more parameters than LLaMA-3 Block: 364 layers added, 2 removed, 5 changed.**

Source: https://neurarch.com/diff/llama3-block-vs-deepseek-v3.html

## Sides

| | LLaMA-3 Block | DeepSeek-V3 |
|---|---|---|
| Layers | 8 | 370 |
| Parameters | 703M | 666B |
| Input | 1 × 2048 | 1 × 163840 |
| Output | 1 × 2048 × 4096 | 1 × 163840 × 129280 |
| Forward-passes | yes | yes |
| Est. train cost | $14.27 | $670,780 |
| T4 16GB | fits | no |
| A100 40GB | fits | no |
| H100 80GB | fits | no |

## Deltas (DeepSeek-V3 relative to LLaMA-3 Block)

- Parameters: +666B (948× the size)
- Layers: +362
- Added 364, removed 2, changed 5, unchanged 3

## Layer by layer

| # | Status | LLaMA-3 Block | Params | Output | DeepSeek-V3 | Params | Output |
|---|---|---|---|---|---|---|---|
| 1 | changed (shape) | tokens (Input) |  | 1 × 2048 | Input (Input) |  | 1 × 163840 |
| 2 | removed | rope (Rope) |  |  |  | |  |
| 3 | changed (numEmbeddings, embeddingDim, vocabSize, maxSeqLen) | embed (Embedding) | 525M | 1 × 2048 × 4096 | Embedding (Embedding) | 927M | 1 × 163840 × 7168 |
| 4 | added |  | |  | RoPE (decoupled) (Rope) |  | 1 × 163840 × 7168 |
| 5 | changed (normalizedShape) | attn_norm (Rms Norm) | 4.1K | 1 × 2048 × 4096 | RMSNorm_1_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 6 | removed | gqa (Grouped Query Attention) | 42M | 1 × 2048 × 4096 |  | |  |
| 7 | added |  | |  | MLA_1 (Mla) | 110M | 1 × 163840 × 7168 |
| 8 | added |  | |  | Add_1_attn (Add) |  | 1 × 163840 × 7168 |
| 9 | added |  | |  | RMSNorm_1_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 10 | added |  | |  | FFN_1 (Swiglu) | 396M | 1 × 163840 × 7168 |
| 11 | added |  | |  | Add_1_ffn (Add) |  | 1 × 163840 × 7168 |
| 12 | added |  | |  | RMSNorm_2_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 13 | added |  | |  | MLA_2 (Mla) | 110M | 1 × 163840 × 7168 |
| 14 | added |  | |  | Add_2_attn (Add) |  | 1 × 163840 × 7168 |
| 15 | added |  | |  | RMSNorm_2_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 16 | added |  | |  | FFN_2 (Swiglu) | 396M | 1 × 163840 × 7168 |
| 17 | added |  | |  | Add_2_ffn (Add) |  | 1 × 163840 × 7168 |
| 18 | added |  | |  | RMSNorm_3_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 19 | added |  | |  | MLA_3 (Mla) | 110M | 1 × 163840 × 7168 |
| 20 | same | residual_1 (Add) |  | 1 × 2048 × 4096 | Add_3_attn (Add) |  | 1 × 163840 × 7168 |
| 21 | changed (normalizedShape) | ffn_norm (Rms Norm) | 4.1K | 1 × 2048 × 4096 | RMSNorm_3_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 22 | changed (embedDim, intermediateSize, dim) | swiglu_ffn (Swiglu) | 135M | 1 × 2048 × 4096 | FFN_3 (Swiglu) | 396M | 1 × 163840 × 7168 |
| 23 | added |  | |  | Add_3_ffn (Add) |  | 1 × 163840 × 7168 |
| 24 | added |  | |  | RMSNorm_4_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 25 | added |  | |  | MLA_4 (Mla) | 110M | 1 × 163840 × 7168 |
| 26 | added |  | |  | Add_4_attn (Add) |  | 1 × 163840 × 7168 |
| 27 | added |  | |  | RMSNorm_4_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 28 | added |  | |  | MoE_4 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 29 | added |  | |  | Add_4_ffn (Add) |  | 1 × 163840 × 7168 |
| 30 | added |  | |  | RMSNorm_5_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 31 | added |  | |  | MLA_5 (Mla) | 110M | 1 × 163840 × 7168 |
| 32 | added |  | |  | Add_5_attn (Add) |  | 1 × 163840 × 7168 |
| 33 | added |  | |  | RMSNorm_5_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 34 | added |  | |  | MoE_5 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 35 | added |  | |  | Add_5_ffn (Add) |  | 1 × 163840 × 7168 |
| 36 | added |  | |  | RMSNorm_6_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 37 | added |  | |  | MLA_6 (Mla) | 110M | 1 × 163840 × 7168 |
| 38 | added |  | |  | Add_6_attn (Add) |  | 1 × 163840 × 7168 |
| 39 | added |  | |  | RMSNorm_6_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 40 | added |  | |  | MoE_6 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 41 | added |  | |  | Add_6_ffn (Add) |  | 1 × 163840 × 7168 |
| 42 | added |  | |  | RMSNorm_7_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 43 | added |  | |  | MLA_7 (Mla) | 110M | 1 × 163840 × 7168 |
| 44 | added |  | |  | Add_7_attn (Add) |  | 1 × 163840 × 7168 |
| 45 | added |  | |  | RMSNorm_7_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 46 | added |  | |  | MoE_7 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 47 | added |  | |  | Add_7_ffn (Add) |  | 1 × 163840 × 7168 |
| 48 | added |  | |  | RMSNorm_8_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 49 | added |  | |  | MLA_8 (Mla) | 110M | 1 × 163840 × 7168 |
| 50 | added |  | |  | Add_8_attn (Add) |  | 1 × 163840 × 7168 |
| 51 | added |  | |  | RMSNorm_8_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 52 | added |  | |  | MoE_8 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 53 | added |  | |  | Add_8_ffn (Add) |  | 1 × 163840 × 7168 |
| 54 | added |  | |  | RMSNorm_9_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 55 | added |  | |  | MLA_9 (Mla) | 110M | 1 × 163840 × 7168 |
| 56 | added |  | |  | Add_9_attn (Add) |  | 1 × 163840 × 7168 |
| 57 | added |  | |  | RMSNorm_9_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 58 | added |  | |  | MoE_9 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 59 | added |  | |  | Add_9_ffn (Add) |  | 1 × 163840 × 7168 |
| 60 | added |  | |  | RMSNorm_10_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 61 | added |  | |  | MLA_10 (Mla) | 110M | 1 × 163840 × 7168 |
| 62 | added |  | |  | Add_10_attn (Add) |  | 1 × 163840 × 7168 |
| 63 | added |  | |  | RMSNorm_10_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 64 | added |  | |  | MoE_10 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 65 | added |  | |  | Add_10_ffn (Add) |  | 1 × 163840 × 7168 |
| 66 | added |  | |  | RMSNorm_11_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 67 | added |  | |  | MLA_11 (Mla) | 110M | 1 × 163840 × 7168 |
| 68 | added |  | |  | Add_11_attn (Add) |  | 1 × 163840 × 7168 |
| 69 | added |  | |  | RMSNorm_11_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 70 | added |  | |  | MoE_11 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 71 | added |  | |  | Add_11_ffn (Add) |  | 1 × 163840 × 7168 |
| 72 | added |  | |  | RMSNorm_12_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 73 | added |  | |  | MLA_12 (Mla) | 110M | 1 × 163840 × 7168 |
| 74 | added |  | |  | Add_12_attn (Add) |  | 1 × 163840 × 7168 |
| 75 | added |  | |  | RMSNorm_12_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 76 | added |  | |  | MoE_12 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 77 | added |  | |  | Add_12_ffn (Add) |  | 1 × 163840 × 7168 |
| 78 | added |  | |  | RMSNorm_13_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 79 | added |  | |  | MLA_13 (Mla) | 110M | 1 × 163840 × 7168 |
| 80 | added |  | |  | Add_13_attn (Add) |  | 1 × 163840 × 7168 |
| 81 | added |  | |  | RMSNorm_13_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 82 | added |  | |  | MoE_13 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 83 | added |  | |  | Add_13_ffn (Add) |  | 1 × 163840 × 7168 |
| 84 | added |  | |  | RMSNorm_14_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 85 | added |  | |  | MLA_14 (Mla) | 110M | 1 × 163840 × 7168 |
| 86 | added |  | |  | Add_14_attn (Add) |  | 1 × 163840 × 7168 |
| 87 | added |  | |  | RMSNorm_14_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 88 | added |  | |  | MoE_14 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 89 | added |  | |  | Add_14_ffn (Add) |  | 1 × 163840 × 7168 |
| 90 | added |  | |  | RMSNorm_15_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 91 | added |  | |  | MLA_15 (Mla) | 110M | 1 × 163840 × 7168 |
| 92 | added |  | |  | Add_15_attn (Add) |  | 1 × 163840 × 7168 |
| 93 | added |  | |  | RMSNorm_15_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 94 | added |  | |  | MoE_15 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 95 | added |  | |  | Add_15_ffn (Add) |  | 1 × 163840 × 7168 |
| 96 | added |  | |  | RMSNorm_16_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 97 | added |  | |  | MLA_16 (Mla) | 110M | 1 × 163840 × 7168 |
| 98 | added |  | |  | Add_16_attn (Add) |  | 1 × 163840 × 7168 |
| 99 | added |  | |  | RMSNorm_16_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 100 | added |  | |  | MoE_16 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 101 | added |  | |  | Add_16_ffn (Add) |  | 1 × 163840 × 7168 |
| 102 | added |  | |  | RMSNorm_17_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 103 | added |  | |  | MLA_17 (Mla) | 110M | 1 × 163840 × 7168 |
| 104 | added |  | |  | Add_17_attn (Add) |  | 1 × 163840 × 7168 |
| 105 | added |  | |  | RMSNorm_17_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 106 | added |  | |  | MoE_17 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 107 | added |  | |  | Add_17_ffn (Add) |  | 1 × 163840 × 7168 |
| 108 | added |  | |  | RMSNorm_18_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 109 | added |  | |  | MLA_18 (Mla) | 110M | 1 × 163840 × 7168 |
| 110 | added |  | |  | Add_18_attn (Add) |  | 1 × 163840 × 7168 |
| 111 | added |  | |  | RMSNorm_18_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 112 | added |  | |  | MoE_18 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 113 | added |  | |  | Add_18_ffn (Add) |  | 1 × 163840 × 7168 |
| 114 | added |  | |  | RMSNorm_19_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 115 | added |  | |  | MLA_19 (Mla) | 110M | 1 × 163840 × 7168 |
| 116 | added |  | |  | Add_19_attn (Add) |  | 1 × 163840 × 7168 |
| 117 | added |  | |  | RMSNorm_19_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 118 | added |  | |  | MoE_19 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 119 | added |  | |  | Add_19_ffn (Add) |  | 1 × 163840 × 7168 |
| 120 | added |  | |  | RMSNorm_20_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 121 | added |  | |  | MLA_20 (Mla) | 110M | 1 × 163840 × 7168 |
| 122 | added |  | |  | Add_20_attn (Add) |  | 1 × 163840 × 7168 |
| 123 | added |  | |  | RMSNorm_20_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 124 | added |  | |  | MoE_20 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 125 | added |  | |  | Add_20_ffn (Add) |  | 1 × 163840 × 7168 |
| 126 | added |  | |  | RMSNorm_21_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 127 | added |  | |  | MLA_21 (Mla) | 110M | 1 × 163840 × 7168 |
| 128 | added |  | |  | Add_21_attn (Add) |  | 1 × 163840 × 7168 |
| 129 | added |  | |  | RMSNorm_21_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 130 | added |  | |  | MoE_21 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 131 | added |  | |  | Add_21_ffn (Add) |  | 1 × 163840 × 7168 |
| 132 | added |  | |  | RMSNorm_22_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 133 | added |  | |  | MLA_22 (Mla) | 110M | 1 × 163840 × 7168 |
| 134 | added |  | |  | Add_22_attn (Add) |  | 1 × 163840 × 7168 |
| 135 | added |  | |  | RMSNorm_22_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 136 | added |  | |  | MoE_22 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 137 | added |  | |  | Add_22_ffn (Add) |  | 1 × 163840 × 7168 |
| 138 | added |  | |  | RMSNorm_23_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 139 | added |  | |  | MLA_23 (Mla) | 110M | 1 × 163840 × 7168 |
| 140 | added |  | |  | Add_23_attn (Add) |  | 1 × 163840 × 7168 |
| 141 | added |  | |  | RMSNorm_23_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 142 | added |  | |  | MoE_23 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 143 | added |  | |  | Add_23_ffn (Add) |  | 1 × 163840 × 7168 |
| 144 | added |  | |  | RMSNorm_24_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 145 | added |  | |  | MLA_24 (Mla) | 110M | 1 × 163840 × 7168 |
| 146 | added |  | |  | Add_24_attn (Add) |  | 1 × 163840 × 7168 |
| 147 | added |  | |  | RMSNorm_24_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 148 | added |  | |  | MoE_24 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 149 | added |  | |  | Add_24_ffn (Add) |  | 1 × 163840 × 7168 |
| 150 | added |  | |  | RMSNorm_25_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 151 | added |  | |  | MLA_25 (Mla) | 110M | 1 × 163840 × 7168 |
| 152 | added |  | |  | Add_25_attn (Add) |  | 1 × 163840 × 7168 |
| 153 | added |  | |  | RMSNorm_25_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 154 | added |  | |  | MoE_25 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 155 | added |  | |  | Add_25_ffn (Add) |  | 1 × 163840 × 7168 |
| 156 | added |  | |  | RMSNorm_26_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 157 | added |  | |  | MLA_26 (Mla) | 110M | 1 × 163840 × 7168 |
| 158 | added |  | |  | Add_26_attn (Add) |  | 1 × 163840 × 7168 |
| 159 | added |  | |  | RMSNorm_26_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 160 | added |  | |  | MoE_26 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 161 | added |  | |  | Add_26_ffn (Add) |  | 1 × 163840 × 7168 |
| 162 | added |  | |  | RMSNorm_27_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 163 | added |  | |  | MLA_27 (Mla) | 110M | 1 × 163840 × 7168 |
| 164 | added |  | |  | Add_27_attn (Add) |  | 1 × 163840 × 7168 |
| 165 | added |  | |  | RMSNorm_27_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 166 | added |  | |  | MoE_27 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 167 | added |  | |  | Add_27_ffn (Add) |  | 1 × 163840 × 7168 |
| 168 | added |  | |  | RMSNorm_28_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 169 | added |  | |  | MLA_28 (Mla) | 110M | 1 × 163840 × 7168 |
| 170 | added |  | |  | Add_28_attn (Add) |  | 1 × 163840 × 7168 |
| 171 | added |  | |  | RMSNorm_28_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 172 | added |  | |  | MoE_28 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 173 | added |  | |  | Add_28_ffn (Add) |  | 1 × 163840 × 7168 |
| 174 | added |  | |  | RMSNorm_29_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 175 | added |  | |  | MLA_29 (Mla) | 110M | 1 × 163840 × 7168 |
| 176 | added |  | |  | Add_29_attn (Add) |  | 1 × 163840 × 7168 |
| 177 | added |  | |  | RMSNorm_29_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 178 | added |  | |  | MoE_29 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 179 | added |  | |  | Add_29_ffn (Add) |  | 1 × 163840 × 7168 |
| 180 | added |  | |  | RMSNorm_30_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 181 | added |  | |  | MLA_30 (Mla) | 110M | 1 × 163840 × 7168 |
| 182 | added |  | |  | Add_30_attn (Add) |  | 1 × 163840 × 7168 |
| 183 | added |  | |  | RMSNorm_30_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 184 | added |  | |  | MoE_30 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 185 | added |  | |  | Add_30_ffn (Add) |  | 1 × 163840 × 7168 |
| 186 | added |  | |  | RMSNorm_31_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 187 | added |  | |  | MLA_31 (Mla) | 110M | 1 × 163840 × 7168 |
| 188 | added |  | |  | Add_31_attn (Add) |  | 1 × 163840 × 7168 |
| 189 | added |  | |  | RMSNorm_31_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 190 | added |  | |  | MoE_31 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 191 | added |  | |  | Add_31_ffn (Add) |  | 1 × 163840 × 7168 |
| 192 | added |  | |  | RMSNorm_32_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 193 | added |  | |  | MLA_32 (Mla) | 110M | 1 × 163840 × 7168 |
| 194 | added |  | |  | Add_32_attn (Add) |  | 1 × 163840 × 7168 |
| 195 | added |  | |  | RMSNorm_32_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 196 | added |  | |  | MoE_32 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 197 | added |  | |  | Add_32_ffn (Add) |  | 1 × 163840 × 7168 |
| 198 | added |  | |  | RMSNorm_33_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 199 | added |  | |  | MLA_33 (Mla) | 110M | 1 × 163840 × 7168 |
| 200 | added |  | |  | Add_33_attn (Add) |  | 1 × 163840 × 7168 |
| 201 | added |  | |  | RMSNorm_33_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 202 | added |  | |  | MoE_33 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 203 | added |  | |  | Add_33_ffn (Add) |  | 1 × 163840 × 7168 |
| 204 | added |  | |  | RMSNorm_34_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 205 | added |  | |  | MLA_34 (Mla) | 110M | 1 × 163840 × 7168 |
| 206 | added |  | |  | Add_34_attn (Add) |  | 1 × 163840 × 7168 |
| 207 | added |  | |  | RMSNorm_34_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 208 | added |  | |  | MoE_34 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 209 | added |  | |  | Add_34_ffn (Add) |  | 1 × 163840 × 7168 |
| 210 | added |  | |  | RMSNorm_35_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 211 | added |  | |  | MLA_35 (Mla) | 110M | 1 × 163840 × 7168 |
| 212 | added |  | |  | Add_35_attn (Add) |  | 1 × 163840 × 7168 |
| 213 | added |  | |  | RMSNorm_35_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 214 | added |  | |  | MoE_35 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 215 | added |  | |  | Add_35_ffn (Add) |  | 1 × 163840 × 7168 |
| 216 | added |  | |  | RMSNorm_36_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 217 | added |  | |  | MLA_36 (Mla) | 110M | 1 × 163840 × 7168 |
| 218 | added |  | |  | Add_36_attn (Add) |  | 1 × 163840 × 7168 |
| 219 | added |  | |  | RMSNorm_36_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 220 | added |  | |  | MoE_36 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 221 | added |  | |  | Add_36_ffn (Add) |  | 1 × 163840 × 7168 |
| 222 | added |  | |  | RMSNorm_37_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 223 | added |  | |  | MLA_37 (Mla) | 110M | 1 × 163840 × 7168 |
| 224 | added |  | |  | Add_37_attn (Add) |  | 1 × 163840 × 7168 |
| 225 | added |  | |  | RMSNorm_37_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 226 | added |  | |  | MoE_37 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 227 | added |  | |  | Add_37_ffn (Add) |  | 1 × 163840 × 7168 |
| 228 | added |  | |  | RMSNorm_38_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 229 | added |  | |  | MLA_38 (Mla) | 110M | 1 × 163840 × 7168 |
| 230 | added |  | |  | Add_38_attn (Add) |  | 1 × 163840 × 7168 |
| 231 | added |  | |  | RMSNorm_38_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 232 | added |  | |  | MoE_38 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 233 | added |  | |  | Add_38_ffn (Add) |  | 1 × 163840 × 7168 |
| 234 | added |  | |  | RMSNorm_39_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 235 | added |  | |  | MLA_39 (Mla) | 110M | 1 × 163840 × 7168 |
| 236 | added |  | |  | Add_39_attn (Add) |  | 1 × 163840 × 7168 |
| 237 | added |  | |  | RMSNorm_39_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 238 | added |  | |  | MoE_39 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 239 | added |  | |  | Add_39_ffn (Add) |  | 1 × 163840 × 7168 |
| 240 | added |  | |  | RMSNorm_40_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 241 | added |  | |  | MLA_40 (Mla) | 110M | 1 × 163840 × 7168 |
| 242 | added |  | |  | Add_40_attn (Add) |  | 1 × 163840 × 7168 |
| 243 | added |  | |  | RMSNorm_40_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 244 | added |  | |  | MoE_40 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 245 | added |  | |  | Add_40_ffn (Add) |  | 1 × 163840 × 7168 |
| 246 | added |  | |  | RMSNorm_41_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 247 | added |  | |  | MLA_41 (Mla) | 110M | 1 × 163840 × 7168 |
| 248 | added |  | |  | Add_41_attn (Add) |  | 1 × 163840 × 7168 |
| 249 | added |  | |  | RMSNorm_41_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 250 | added |  | |  | MoE_41 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 251 | added |  | |  | Add_41_ffn (Add) |  | 1 × 163840 × 7168 |
| 252 | added |  | |  | RMSNorm_42_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 253 | added |  | |  | MLA_42 (Mla) | 110M | 1 × 163840 × 7168 |
| 254 | added |  | |  | Add_42_attn (Add) |  | 1 × 163840 × 7168 |
| 255 | added |  | |  | RMSNorm_42_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 256 | added |  | |  | MoE_42 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 257 | added |  | |  | Add_42_ffn (Add) |  | 1 × 163840 × 7168 |
| 258 | added |  | |  | RMSNorm_43_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 259 | added |  | |  | MLA_43 (Mla) | 110M | 1 × 163840 × 7168 |
| 260 | added |  | |  | Add_43_attn (Add) |  | 1 × 163840 × 7168 |
| 261 | added |  | |  | RMSNorm_43_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 262 | added |  | |  | MoE_43 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 263 | added |  | |  | Add_43_ffn (Add) |  | 1 × 163840 × 7168 |
| 264 | added |  | |  | RMSNorm_44_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 265 | added |  | |  | MLA_44 (Mla) | 110M | 1 × 163840 × 7168 |
| 266 | added |  | |  | Add_44_attn (Add) |  | 1 × 163840 × 7168 |
| 267 | added |  | |  | RMSNorm_44_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 268 | added |  | |  | MoE_44 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 269 | added |  | |  | Add_44_ffn (Add) |  | 1 × 163840 × 7168 |
| 270 | added |  | |  | RMSNorm_45_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 271 | added |  | |  | MLA_45 (Mla) | 110M | 1 × 163840 × 7168 |
| 272 | added |  | |  | Add_45_attn (Add) |  | 1 × 163840 × 7168 |
| 273 | added |  | |  | RMSNorm_45_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 274 | added |  | |  | MoE_45 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 275 | added |  | |  | Add_45_ffn (Add) |  | 1 × 163840 × 7168 |
| 276 | added |  | |  | RMSNorm_46_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 277 | added |  | |  | MLA_46 (Mla) | 110M | 1 × 163840 × 7168 |
| 278 | added |  | |  | Add_46_attn (Add) |  | 1 × 163840 × 7168 |
| 279 | added |  | |  | RMSNorm_46_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 280 | added |  | |  | MoE_46 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 281 | added |  | |  | Add_46_ffn (Add) |  | 1 × 163840 × 7168 |
| 282 | added |  | |  | RMSNorm_47_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 283 | added |  | |  | MLA_47 (Mla) | 110M | 1 × 163840 × 7168 |
| 284 | added |  | |  | Add_47_attn (Add) |  | 1 × 163840 × 7168 |
| 285 | added |  | |  | RMSNorm_47_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 286 | added |  | |  | MoE_47 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 287 | added |  | |  | Add_47_ffn (Add) |  | 1 × 163840 × 7168 |
| 288 | added |  | |  | RMSNorm_48_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 289 | added |  | |  | MLA_48 (Mla) | 110M | 1 × 163840 × 7168 |
| 290 | added |  | |  | Add_48_attn (Add) |  | 1 × 163840 × 7168 |
| 291 | added |  | |  | RMSNorm_48_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 292 | added |  | |  | MoE_48 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 293 | added |  | |  | Add_48_ffn (Add) |  | 1 × 163840 × 7168 |
| 294 | added |  | |  | RMSNorm_49_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 295 | added |  | |  | MLA_49 (Mla) | 110M | 1 × 163840 × 7168 |
| 296 | added |  | |  | Add_49_attn (Add) |  | 1 × 163840 × 7168 |
| 297 | added |  | |  | RMSNorm_49_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 298 | added |  | |  | MoE_49 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 299 | added |  | |  | Add_49_ffn (Add) |  | 1 × 163840 × 7168 |
| 300 | added |  | |  | RMSNorm_50_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 301 | added |  | |  | MLA_50 (Mla) | 110M | 1 × 163840 × 7168 |
| 302 | added |  | |  | Add_50_attn (Add) |  | 1 × 163840 × 7168 |
| 303 | added |  | |  | RMSNorm_50_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 304 | added |  | |  | MoE_50 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 305 | added |  | |  | Add_50_ffn (Add) |  | 1 × 163840 × 7168 |
| 306 | added |  | |  | RMSNorm_51_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 307 | added |  | |  | MLA_51 (Mla) | 110M | 1 × 163840 × 7168 |
| 308 | added |  | |  | Add_51_attn (Add) |  | 1 × 163840 × 7168 |
| 309 | added |  | |  | RMSNorm_51_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 310 | added |  | |  | MoE_51 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 311 | added |  | |  | Add_51_ffn (Add) |  | 1 × 163840 × 7168 |
| 312 | added |  | |  | RMSNorm_52_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 313 | added |  | |  | MLA_52 (Mla) | 110M | 1 × 163840 × 7168 |
| 314 | added |  | |  | Add_52_attn (Add) |  | 1 × 163840 × 7168 |
| 315 | added |  | |  | RMSNorm_52_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 316 | added |  | |  | MoE_52 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 317 | added |  | |  | Add_52_ffn (Add) |  | 1 × 163840 × 7168 |
| 318 | added |  | |  | RMSNorm_53_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 319 | added |  | |  | MLA_53 (Mla) | 110M | 1 × 163840 × 7168 |
| 320 | added |  | |  | Add_53_attn (Add) |  | 1 × 163840 × 7168 |
| 321 | added |  | |  | RMSNorm_53_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 322 | added |  | |  | MoE_53 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 323 | added |  | |  | Add_53_ffn (Add) |  | 1 × 163840 × 7168 |
| 324 | added |  | |  | RMSNorm_54_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 325 | added |  | |  | MLA_54 (Mla) | 110M | 1 × 163840 × 7168 |
| 326 | added |  | |  | Add_54_attn (Add) |  | 1 × 163840 × 7168 |
| 327 | added |  | |  | RMSNorm_54_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 328 | added |  | |  | MoE_54 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 329 | added |  | |  | Add_54_ffn (Add) |  | 1 × 163840 × 7168 |
| 330 | added |  | |  | RMSNorm_55_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 331 | added |  | |  | MLA_55 (Mla) | 110M | 1 × 163840 × 7168 |
| 332 | added |  | |  | Add_55_attn (Add) |  | 1 × 163840 × 7168 |
| 333 | added |  | |  | RMSNorm_55_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 334 | added |  | |  | MoE_55 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 335 | added |  | |  | Add_55_ffn (Add) |  | 1 × 163840 × 7168 |
| 336 | added |  | |  | RMSNorm_56_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 337 | added |  | |  | MLA_56 (Mla) | 110M | 1 × 163840 × 7168 |
| 338 | added |  | |  | Add_56_attn (Add) |  | 1 × 163840 × 7168 |
| 339 | added |  | |  | RMSNorm_56_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 340 | added |  | |  | MoE_56 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 341 | added |  | |  | Add_56_ffn (Add) |  | 1 × 163840 × 7168 |
| 342 | added |  | |  | RMSNorm_57_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 343 | added |  | |  | MLA_57 (Mla) | 110M | 1 × 163840 × 7168 |
| 344 | added |  | |  | Add_57_attn (Add) |  | 1 × 163840 × 7168 |
| 345 | added |  | |  | RMSNorm_57_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 346 | added |  | |  | MoE_57 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 347 | added |  | |  | Add_57_ffn (Add) |  | 1 × 163840 × 7168 |
| 348 | added |  | |  | RMSNorm_58_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 349 | added |  | |  | MLA_58 (Mla) | 110M | 1 × 163840 × 7168 |
| 350 | added |  | |  | Add_58_attn (Add) |  | 1 × 163840 × 7168 |
| 351 | added |  | |  | RMSNorm_58_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 352 | added |  | |  | MoE_58 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 353 | added |  | |  | Add_58_ffn (Add) |  | 1 × 163840 × 7168 |
| 354 | added |  | |  | RMSNorm_59_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 355 | added |  | |  | MLA_59 (Mla) | 110M | 1 × 163840 × 7168 |
| 356 | added |  | |  | Add_59_attn (Add) |  | 1 × 163840 × 7168 |
| 357 | added |  | |  | RMSNorm_59_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 358 | added |  | |  | MoE_59 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 359 | added |  | |  | Add_59_ffn (Add) |  | 1 × 163840 × 7168 |
| 360 | added |  | |  | RMSNorm_60_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 361 | added |  | |  | MLA_60 (Mla) | 110M | 1 × 163840 × 7168 |
| 362 | added |  | |  | Add_60_attn (Add) |  | 1 × 163840 × 7168 |
| 363 | added |  | |  | RMSNorm_60_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 364 | added |  | |  | MoE_60 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 365 | added |  | |  | Add_60_ffn (Add) |  | 1 × 163840 × 7168 |
| 366 | added |  | |  | RMSNorm_61_1 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 367 | added |  | |  | MLA_61 (Mla) | 110M | 1 × 163840 × 7168 |
| 368 | added |  | |  | Add_61_attn (Add) |  | 1 × 163840 × 7168 |
| 369 | added |  | |  | RMSNorm_61_2 (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 370 | added |  | |  | MoE_61 (Shared Expert Mo E) | 11B | 1 × 163840 × 7168 |
| 371 | same | residual_2 (Add) |  | 1 × 2048 × 4096 | Add_61_ffn (Add) |  | 1 × 163840 × 7168 |
| 372 | added |  | |  | Final_RMSNorm (Rms Norm) | 7.2K | 1 × 163840 × 7168 |
| 373 | added |  | |  | LM_Head (Linear) | 927M | 1 × 163840 × 129280 |
| 374 | same | hidden_state (Output) |  | 1 × 2048 × 4096 | Output (Output) |  | 1 × 163840 × 129280 |

## What this is not

- The two are priced at different declared inputs (1 × 2048 against 1 × 163840), so memory, cost and GPU fit are each right about their own model and are not a comparison between them. The layer and parameter deltas are unaffected.
- Parameter counts are derived from the graph, not read from a checkpoint. They are exact for a graph that is fully specified and approximate for one that is not.
- Cost and GPU fit are estimates from the graph under one set of assumptions, not measurements of a run.

## Graphs

- LLaMA-3 Block: https://neurarch.com/templates/llama3-block/model.json
- DeepSeek-V3: https://neurarch.com/templates/deepseek-v3/model.json
- Check a graph of your own: `POST https://www.neurarch.com/api/v1/plan`
