N Neurarch Architectures Models Checks Data Docs Open the app

Comparisons / Mixtral MoE Block vs DeepSeek-V3

Mixtral MoE Block vs DeepSeek-V3

Two mixture-of-experts designs that route very differently.

DeepSeek-V3 has 665B more parameters than Mixtral MoE Block: 365 layers added, 2 removed, 4 changed.

Baseline

Mixtral MoE Block

Layers
7
Parameters
1.5B
Input
1 × 4096 × 4096
Output
1 × 4096 × 4096
Forward-passes
yes
Est. train cost
$63.07
Compared

DeepSeek-V3

Layers
370
Parameters
666B
Input
1 × 163840
Output
1 × 163840 × 129280
Forward-passes
yes
Est. train cost
$670,780

The deltas

Every number is DeepSeek-V3 relative to Mixtral MoE Block.

Parameters
+665B (459× the size)
Layers
+363
Added
365
Removed
2
Changed
4
Unchanged
3

Which GPUs each one fits

Each side is measured at its own declared input (1 × 4096 × 4096 against 1 × 163840). Both columns are right about their own model; the difference between them is not a fact about the designs.

GPUMixtral MoE BlockDeepSeek-V3
T4 16GBnono
A100 40GBfitsno
H100 80GBfitsno

Layer by layer

Aligned in topological order. 3 of 374 rows are the same layer with the same parameters.

Show all 374 rows
Mixtral MoE BlockDeepSeek-V3
LayerParamsOutputLayerParamsOutput
1changed
shape
hidden_states
Input
1 × 4096 × 4096Input
Input
1 × 163840
2addedEmbedding
Embedding
927M1 × 163840 × 7168
3changed
dim, maxPositionEmbeddings
rotary_emb
Rope
RoPE (decoupled)
Rope
1 × 163840 × 7168
4changed
normalizedShape
input_norm
Rms Norm
4.1K1 × 4096 × 4096RMSNorm_1_1
Rms Norm
7.2K1 × 163840 × 7168
5removedself_attn
Grouped Query Attention
42M1 × 4096 × 4096
6addedMLA_1
Mla
110M1 × 163840 × 7168
7sameattn_residual
Add
1 × 4096 × 4096Add_1_attn
Add
1 × 163840 × 7168
8changed
normalizedShape
post_attn_norm
Rms Norm
4.1K1 × 4096 × 4096RMSNorm_1_2
Rms Norm
7.2K1 × 163840 × 7168
9removedblock_sparse_moe
Moe Layer
1.4B1 × 4096 × 4096
10addedFFN_1
Swiglu
396M1 × 163840 × 7168
11addedAdd_1_ffn
Add
1 × 163840 × 7168
12addedRMSNorm_2_1
Rms Norm
7.2K1 × 163840 × 7168
13addedMLA_2
Mla
110M1 × 163840 × 7168
14addedAdd_2_attn
Add
1 × 163840 × 7168
15addedRMSNorm_2_2
Rms Norm
7.2K1 × 163840 × 7168
16addedFFN_2
Swiglu
396M1 × 163840 × 7168
17addedAdd_2_ffn
Add
1 × 163840 × 7168
18addedRMSNorm_3_1
Rms Norm
7.2K1 × 163840 × 7168
19addedMLA_3
Mla
110M1 × 163840 × 7168
20addedAdd_3_attn
Add
1 × 163840 × 7168
21addedRMSNorm_3_2
Rms Norm
7.2K1 × 163840 × 7168
22addedFFN_3
Swiglu
396M1 × 163840 × 7168
23addedAdd_3_ffn
Add
1 × 163840 × 7168
24addedRMSNorm_4_1
Rms Norm
7.2K1 × 163840 × 7168
25addedMLA_4
Mla
110M1 × 163840 × 7168
26addedAdd_4_attn
Add
1 × 163840 × 7168
27addedRMSNorm_4_2
Rms Norm
7.2K1 × 163840 × 7168
28addedMoE_4
Shared Expert Mo E
11B1 × 163840 × 7168
29addedAdd_4_ffn
Add
1 × 163840 × 7168
30addedRMSNorm_5_1
Rms Norm
7.2K1 × 163840 × 7168
31addedMLA_5
Mla
110M1 × 163840 × 7168
32addedAdd_5_attn
Add
1 × 163840 × 7168
33addedRMSNorm_5_2
Rms Norm
7.2K1 × 163840 × 7168
34addedMoE_5
Shared Expert Mo E
11B1 × 163840 × 7168
35addedAdd_5_ffn
Add
1 × 163840 × 7168
36addedRMSNorm_6_1
Rms Norm
7.2K1 × 163840 × 7168
37addedMLA_6
Mla
110M1 × 163840 × 7168
38addedAdd_6_attn
Add
1 × 163840 × 7168
39addedRMSNorm_6_2
Rms Norm
7.2K1 × 163840 × 7168
40addedMoE_6
Shared Expert Mo E
11B1 × 163840 × 7168
41addedAdd_6_ffn
Add
1 × 163840 × 7168
42addedRMSNorm_7_1
Rms Norm
7.2K1 × 163840 × 7168
43addedMLA_7
Mla
110M1 × 163840 × 7168
44addedAdd_7_attn
Add
1 × 163840 × 7168
45addedRMSNorm_7_2
Rms Norm
7.2K1 × 163840 × 7168
46addedMoE_7
Shared Expert Mo E
11B1 × 163840 × 7168
47addedAdd_7_ffn
Add
1 × 163840 × 7168
48addedRMSNorm_8_1
Rms Norm
7.2K1 × 163840 × 7168
49addedMLA_8
Mla
110M1 × 163840 × 7168
50addedAdd_8_attn
Add
1 × 163840 × 7168
51addedRMSNorm_8_2
Rms Norm
7.2K1 × 163840 × 7168
52addedMoE_8
Shared Expert Mo E
11B1 × 163840 × 7168
53addedAdd_8_ffn
Add
1 × 163840 × 7168
54addedRMSNorm_9_1
Rms Norm
7.2K1 × 163840 × 7168
55addedMLA_9
Mla
110M1 × 163840 × 7168
56addedAdd_9_attn
Add
1 × 163840 × 7168
57addedRMSNorm_9_2
Rms Norm
7.2K1 × 163840 × 7168
58addedMoE_9
Shared Expert Mo E
11B1 × 163840 × 7168
59addedAdd_9_ffn
Add
1 × 163840 × 7168
60addedRMSNorm_10_1
Rms Norm
7.2K1 × 163840 × 7168
61addedMLA_10
Mla
110M1 × 163840 × 7168
62addedAdd_10_attn
Add
1 × 163840 × 7168
63addedRMSNorm_10_2
Rms Norm
7.2K1 × 163840 × 7168
64addedMoE_10
Shared Expert Mo E
11B1 × 163840 × 7168
65addedAdd_10_ffn
Add
1 × 163840 × 7168
66addedRMSNorm_11_1
Rms Norm
7.2K1 × 163840 × 7168
67addedMLA_11
Mla
110M1 × 163840 × 7168
68addedAdd_11_attn
Add
1 × 163840 × 7168
69addedRMSNorm_11_2
Rms Norm
7.2K1 × 163840 × 7168
70addedMoE_11
Shared Expert Mo E
11B1 × 163840 × 7168
71addedAdd_11_ffn
Add
1 × 163840 × 7168
72addedRMSNorm_12_1
Rms Norm
7.2K1 × 163840 × 7168
73addedMLA_12
Mla
110M1 × 163840 × 7168
74addedAdd_12_attn
Add
1 × 163840 × 7168
75addedRMSNorm_12_2
Rms Norm
7.2K1 × 163840 × 7168
76addedMoE_12
Shared Expert Mo E
11B1 × 163840 × 7168
77addedAdd_12_ffn
Add
1 × 163840 × 7168
78addedRMSNorm_13_1
Rms Norm
7.2K1 × 163840 × 7168
79addedMLA_13
Mla
110M1 × 163840 × 7168
80addedAdd_13_attn
Add
1 × 163840 × 7168
81addedRMSNorm_13_2
Rms Norm
7.2K1 × 163840 × 7168
82addedMoE_13
Shared Expert Mo E
11B1 × 163840 × 7168
83addedAdd_13_ffn
Add
1 × 163840 × 7168
84addedRMSNorm_14_1
Rms Norm
7.2K1 × 163840 × 7168
85addedMLA_14
Mla
110M1 × 163840 × 7168
86addedAdd_14_attn
Add
1 × 163840 × 7168
87addedRMSNorm_14_2
Rms Norm
7.2K1 × 163840 × 7168
88addedMoE_14
Shared Expert Mo E
11B1 × 163840 × 7168
89addedAdd_14_ffn
Add
1 × 163840 × 7168
90addedRMSNorm_15_1
Rms Norm
7.2K1 × 163840 × 7168
91addedMLA_15
Mla
110M1 × 163840 × 7168
92addedAdd_15_attn
Add
1 × 163840 × 7168
93addedRMSNorm_15_2
Rms Norm
7.2K1 × 163840 × 7168
94addedMoE_15
Shared Expert Mo E
11B1 × 163840 × 7168
95addedAdd_15_ffn
Add
1 × 163840 × 7168
96addedRMSNorm_16_1
Rms Norm
7.2K1 × 163840 × 7168
97addedMLA_16
Mla
110M1 × 163840 × 7168
98addedAdd_16_attn
Add
1 × 163840 × 7168
99addedRMSNorm_16_2
Rms Norm
7.2K1 × 163840 × 7168
100addedMoE_16
Shared Expert Mo E
11B1 × 163840 × 7168
101addedAdd_16_ffn
Add
1 × 163840 × 7168
102addedRMSNorm_17_1
Rms Norm
7.2K1 × 163840 × 7168
103addedMLA_17
Mla
110M1 × 163840 × 7168
104addedAdd_17_attn
Add
1 × 163840 × 7168
105addedRMSNorm_17_2
Rms Norm
7.2K1 × 163840 × 7168
106addedMoE_17
Shared Expert Mo E
11B1 × 163840 × 7168
107addedAdd_17_ffn
Add
1 × 163840 × 7168
108addedRMSNorm_18_1
Rms Norm
7.2K1 × 163840 × 7168
109addedMLA_18
Mla
110M1 × 163840 × 7168
110addedAdd_18_attn
Add
1 × 163840 × 7168
111addedRMSNorm_18_2
Rms Norm
7.2K1 × 163840 × 7168
112addedMoE_18
Shared Expert Mo E
11B1 × 163840 × 7168
113addedAdd_18_ffn
Add
1 × 163840 × 7168
114addedRMSNorm_19_1
Rms Norm
7.2K1 × 163840 × 7168
115addedMLA_19
Mla
110M1 × 163840 × 7168
116addedAdd_19_attn
Add
1 × 163840 × 7168
117addedRMSNorm_19_2
Rms Norm
7.2K1 × 163840 × 7168
118addedMoE_19
Shared Expert Mo E
11B1 × 163840 × 7168
119addedAdd_19_ffn
Add
1 × 163840 × 7168
120addedRMSNorm_20_1
Rms Norm
7.2K1 × 163840 × 7168
121addedMLA_20
Mla
110M1 × 163840 × 7168
122addedAdd_20_attn
Add
1 × 163840 × 7168
123addedRMSNorm_20_2
Rms Norm
7.2K1 × 163840 × 7168
124addedMoE_20
Shared Expert Mo E
11B1 × 163840 × 7168
125addedAdd_20_ffn
Add
1 × 163840 × 7168
126addedRMSNorm_21_1
Rms Norm
7.2K1 × 163840 × 7168
127addedMLA_21
Mla
110M1 × 163840 × 7168
128addedAdd_21_attn
Add
1 × 163840 × 7168
129addedRMSNorm_21_2
Rms Norm
7.2K1 × 163840 × 7168
130addedMoE_21
Shared Expert Mo E
11B1 × 163840 × 7168
131addedAdd_21_ffn
Add
1 × 163840 × 7168
132addedRMSNorm_22_1
Rms Norm
7.2K1 × 163840 × 7168
133addedMLA_22
Mla
110M1 × 163840 × 7168
134addedAdd_22_attn
Add
1 × 163840 × 7168
135addedRMSNorm_22_2
Rms Norm
7.2K1 × 163840 × 7168
136addedMoE_22
Shared Expert Mo E
11B1 × 163840 × 7168
137addedAdd_22_ffn
Add
1 × 163840 × 7168
138addedRMSNorm_23_1
Rms Norm
7.2K1 × 163840 × 7168
139addedMLA_23
Mla
110M1 × 163840 × 7168
140addedAdd_23_attn
Add
1 × 163840 × 7168
141addedRMSNorm_23_2
Rms Norm
7.2K1 × 163840 × 7168
142addedMoE_23
Shared Expert Mo E
11B1 × 163840 × 7168
143addedAdd_23_ffn
Add
1 × 163840 × 7168
144addedRMSNorm_24_1
Rms Norm
7.2K1 × 163840 × 7168
145addedMLA_24
Mla
110M1 × 163840 × 7168
146addedAdd_24_attn
Add
1 × 163840 × 7168
147addedRMSNorm_24_2
Rms Norm
7.2K1 × 163840 × 7168
148addedMoE_24
Shared Expert Mo E
11B1 × 163840 × 7168
149addedAdd_24_ffn
Add
1 × 163840 × 7168
150addedRMSNorm_25_1
Rms Norm
7.2K1 × 163840 × 7168
151addedMLA_25
Mla
110M1 × 163840 × 7168
152addedAdd_25_attn
Add
1 × 163840 × 7168
153addedRMSNorm_25_2
Rms Norm
7.2K1 × 163840 × 7168
154addedMoE_25
Shared Expert Mo E
11B1 × 163840 × 7168
155addedAdd_25_ffn
Add
1 × 163840 × 7168
156addedRMSNorm_26_1
Rms Norm
7.2K1 × 163840 × 7168
157addedMLA_26
Mla
110M1 × 163840 × 7168
158addedAdd_26_attn
Add
1 × 163840 × 7168
159addedRMSNorm_26_2
Rms Norm
7.2K1 × 163840 × 7168
160addedMoE_26
Shared Expert Mo E
11B1 × 163840 × 7168
161addedAdd_26_ffn
Add
1 × 163840 × 7168
162addedRMSNorm_27_1
Rms Norm
7.2K1 × 163840 × 7168
163addedMLA_27
Mla
110M1 × 163840 × 7168
164addedAdd_27_attn
Add
1 × 163840 × 7168
165addedRMSNorm_27_2
Rms Norm
7.2K1 × 163840 × 7168
166addedMoE_27
Shared Expert Mo E
11B1 × 163840 × 7168
167addedAdd_27_ffn
Add
1 × 163840 × 7168
168addedRMSNorm_28_1
Rms Norm
7.2K1 × 163840 × 7168
169addedMLA_28
Mla
110M1 × 163840 × 7168
170addedAdd_28_attn
Add
1 × 163840 × 7168
171addedRMSNorm_28_2
Rms Norm
7.2K1 × 163840 × 7168
172addedMoE_28
Shared Expert Mo E
11B1 × 163840 × 7168
173addedAdd_28_ffn
Add
1 × 163840 × 7168
174addedRMSNorm_29_1
Rms Norm
7.2K1 × 163840 × 7168
175addedMLA_29
Mla
110M1 × 163840 × 7168
176addedAdd_29_attn
Add
1 × 163840 × 7168
177addedRMSNorm_29_2
Rms Norm
7.2K1 × 163840 × 7168
178addedMoE_29
Shared Expert Mo E
11B1 × 163840 × 7168
179addedAdd_29_ffn
Add
1 × 163840 × 7168
180addedRMSNorm_30_1
Rms Norm
7.2K1 × 163840 × 7168
181addedMLA_30
Mla
110M1 × 163840 × 7168
182addedAdd_30_attn
Add
1 × 163840 × 7168
183addedRMSNorm_30_2
Rms Norm
7.2K1 × 163840 × 7168
184addedMoE_30
Shared Expert Mo E
11B1 × 163840 × 7168
185addedAdd_30_ffn
Add
1 × 163840 × 7168
186addedRMSNorm_31_1
Rms Norm
7.2K1 × 163840 × 7168
187addedMLA_31
Mla
110M1 × 163840 × 7168
188addedAdd_31_attn
Add
1 × 163840 × 7168
189addedRMSNorm_31_2
Rms Norm
7.2K1 × 163840 × 7168
190addedMoE_31
Shared Expert Mo E
11B1 × 163840 × 7168
191addedAdd_31_ffn
Add
1 × 163840 × 7168
192addedRMSNorm_32_1
Rms Norm
7.2K1 × 163840 × 7168
193addedMLA_32
Mla
110M1 × 163840 × 7168
194addedAdd_32_attn
Add
1 × 163840 × 7168
195addedRMSNorm_32_2
Rms Norm
7.2K1 × 163840 × 7168
196addedMoE_32
Shared Expert Mo E
11B1 × 163840 × 7168
197addedAdd_32_ffn
Add
1 × 163840 × 7168
198addedRMSNorm_33_1
Rms Norm
7.2K1 × 163840 × 7168
199addedMLA_33
Mla
110M1 × 163840 × 7168
200addedAdd_33_attn
Add
1 × 163840 × 7168
201addedRMSNorm_33_2
Rms Norm
7.2K1 × 163840 × 7168
202addedMoE_33
Shared Expert Mo E
11B1 × 163840 × 7168
203addedAdd_33_ffn
Add
1 × 163840 × 7168
204addedRMSNorm_34_1
Rms Norm
7.2K1 × 163840 × 7168
205addedMLA_34
Mla
110M1 × 163840 × 7168
206addedAdd_34_attn
Add
1 × 163840 × 7168
207addedRMSNorm_34_2
Rms Norm
7.2K1 × 163840 × 7168
208addedMoE_34
Shared Expert Mo E
11B1 × 163840 × 7168
209addedAdd_34_ffn
Add
1 × 163840 × 7168
210addedRMSNorm_35_1
Rms Norm
7.2K1 × 163840 × 7168
211addedMLA_35
Mla
110M1 × 163840 × 7168
212addedAdd_35_attn
Add
1 × 163840 × 7168
213addedRMSNorm_35_2
Rms Norm
7.2K1 × 163840 × 7168
214addedMoE_35
Shared Expert Mo E
11B1 × 163840 × 7168
215addedAdd_35_ffn
Add
1 × 163840 × 7168
216addedRMSNorm_36_1
Rms Norm
7.2K1 × 163840 × 7168
217addedMLA_36
Mla
110M1 × 163840 × 7168
218addedAdd_36_attn
Add
1 × 163840 × 7168
219addedRMSNorm_36_2
Rms Norm
7.2K1 × 163840 × 7168
220addedMoE_36
Shared Expert Mo E
11B1 × 163840 × 7168
221addedAdd_36_ffn
Add
1 × 163840 × 7168
222addedRMSNorm_37_1
Rms Norm
7.2K1 × 163840 × 7168
223addedMLA_37
Mla
110M1 × 163840 × 7168
224addedAdd_37_attn
Add
1 × 163840 × 7168
225addedRMSNorm_37_2
Rms Norm
7.2K1 × 163840 × 7168
226addedMoE_37
Shared Expert Mo E
11B1 × 163840 × 7168
227addedAdd_37_ffn
Add
1 × 163840 × 7168
228addedRMSNorm_38_1
Rms Norm
7.2K1 × 163840 × 7168
229addedMLA_38
Mla
110M1 × 163840 × 7168
230addedAdd_38_attn
Add
1 × 163840 × 7168
231addedRMSNorm_38_2
Rms Norm
7.2K1 × 163840 × 7168
232addedMoE_38
Shared Expert Mo E
11B1 × 163840 × 7168
233addedAdd_38_ffn
Add
1 × 163840 × 7168
234addedRMSNorm_39_1
Rms Norm
7.2K1 × 163840 × 7168
235addedMLA_39
Mla
110M1 × 163840 × 7168
236addedAdd_39_attn
Add
1 × 163840 × 7168
237addedRMSNorm_39_2
Rms Norm
7.2K1 × 163840 × 7168
238addedMoE_39
Shared Expert Mo E
11B1 × 163840 × 7168
239addedAdd_39_ffn
Add
1 × 163840 × 7168
240addedRMSNorm_40_1
Rms Norm
7.2K1 × 163840 × 7168
241addedMLA_40
Mla
110M1 × 163840 × 7168
242addedAdd_40_attn
Add
1 × 163840 × 7168
243addedRMSNorm_40_2
Rms Norm
7.2K1 × 163840 × 7168
244addedMoE_40
Shared Expert Mo E
11B1 × 163840 × 7168
245addedAdd_40_ffn
Add
1 × 163840 × 7168
246addedRMSNorm_41_1
Rms Norm
7.2K1 × 163840 × 7168
247addedMLA_41
Mla
110M1 × 163840 × 7168
248addedAdd_41_attn
Add
1 × 163840 × 7168
249addedRMSNorm_41_2
Rms Norm
7.2K1 × 163840 × 7168
250addedMoE_41
Shared Expert Mo E
11B1 × 163840 × 7168
251addedAdd_41_ffn
Add
1 × 163840 × 7168
252addedRMSNorm_42_1
Rms Norm
7.2K1 × 163840 × 7168
253addedMLA_42
Mla
110M1 × 163840 × 7168
254addedAdd_42_attn
Add
1 × 163840 × 7168
255addedRMSNorm_42_2
Rms Norm
7.2K1 × 163840 × 7168
256addedMoE_42
Shared Expert Mo E
11B1 × 163840 × 7168
257addedAdd_42_ffn
Add
1 × 163840 × 7168
258addedRMSNorm_43_1
Rms Norm
7.2K1 × 163840 × 7168
259addedMLA_43
Mla
110M1 × 163840 × 7168
260addedAdd_43_attn
Add
1 × 163840 × 7168
261addedRMSNorm_43_2
Rms Norm
7.2K1 × 163840 × 7168
262addedMoE_43
Shared Expert Mo E
11B1 × 163840 × 7168
263addedAdd_43_ffn
Add
1 × 163840 × 7168
264addedRMSNorm_44_1
Rms Norm
7.2K1 × 163840 × 7168
265addedMLA_44
Mla
110M1 × 163840 × 7168
266addedAdd_44_attn
Add
1 × 163840 × 7168
267addedRMSNorm_44_2
Rms Norm
7.2K1 × 163840 × 7168
268addedMoE_44
Shared Expert Mo E
11B1 × 163840 × 7168
269addedAdd_44_ffn
Add
1 × 163840 × 7168
270addedRMSNorm_45_1
Rms Norm
7.2K1 × 163840 × 7168
271addedMLA_45
Mla
110M1 × 163840 × 7168
272addedAdd_45_attn
Add
1 × 163840 × 7168
273addedRMSNorm_45_2
Rms Norm
7.2K1 × 163840 × 7168
274addedMoE_45
Shared Expert Mo E
11B1 × 163840 × 7168
275addedAdd_45_ffn
Add
1 × 163840 × 7168
276addedRMSNorm_46_1
Rms Norm
7.2K1 × 163840 × 7168
277addedMLA_46
Mla
110M1 × 163840 × 7168
278addedAdd_46_attn
Add
1 × 163840 × 7168
279addedRMSNorm_46_2
Rms Norm
7.2K1 × 163840 × 7168
280addedMoE_46
Shared Expert Mo E
11B1 × 163840 × 7168
281addedAdd_46_ffn
Add
1 × 163840 × 7168
282addedRMSNorm_47_1
Rms Norm
7.2K1 × 163840 × 7168
283addedMLA_47
Mla
110M1 × 163840 × 7168
284addedAdd_47_attn
Add
1 × 163840 × 7168
285addedRMSNorm_47_2
Rms Norm
7.2K1 × 163840 × 7168
286addedMoE_47
Shared Expert Mo E
11B1 × 163840 × 7168
287addedAdd_47_ffn
Add
1 × 163840 × 7168
288addedRMSNorm_48_1
Rms Norm
7.2K1 × 163840 × 7168
289addedMLA_48
Mla
110M1 × 163840 × 7168
290addedAdd_48_attn
Add
1 × 163840 × 7168
291addedRMSNorm_48_2
Rms Norm
7.2K1 × 163840 × 7168
292addedMoE_48
Shared Expert Mo E
11B1 × 163840 × 7168
293addedAdd_48_ffn
Add
1 × 163840 × 7168
294addedRMSNorm_49_1
Rms Norm
7.2K1 × 163840 × 7168
295addedMLA_49
Mla
110M1 × 163840 × 7168
296addedAdd_49_attn
Add
1 × 163840 × 7168
297addedRMSNorm_49_2
Rms Norm
7.2K1 × 163840 × 7168
298addedMoE_49
Shared Expert Mo E
11B1 × 163840 × 7168
299addedAdd_49_ffn
Add
1 × 163840 × 7168
300addedRMSNorm_50_1
Rms Norm
7.2K1 × 163840 × 7168
301addedMLA_50
Mla
110M1 × 163840 × 7168
302addedAdd_50_attn
Add
1 × 163840 × 7168
303addedRMSNorm_50_2
Rms Norm
7.2K1 × 163840 × 7168
304addedMoE_50
Shared Expert Mo E
11B1 × 163840 × 7168
305addedAdd_50_ffn
Add
1 × 163840 × 7168
306addedRMSNorm_51_1
Rms Norm
7.2K1 × 163840 × 7168
307addedMLA_51
Mla
110M1 × 163840 × 7168
308addedAdd_51_attn
Add
1 × 163840 × 7168
309addedRMSNorm_51_2
Rms Norm
7.2K1 × 163840 × 7168
310addedMoE_51
Shared Expert Mo E
11B1 × 163840 × 7168
311addedAdd_51_ffn
Add
1 × 163840 × 7168
312addedRMSNorm_52_1
Rms Norm
7.2K1 × 163840 × 7168
313addedMLA_52
Mla
110M1 × 163840 × 7168
314addedAdd_52_attn
Add
1 × 163840 × 7168
315addedRMSNorm_52_2
Rms Norm
7.2K1 × 163840 × 7168
316addedMoE_52
Shared Expert Mo E
11B1 × 163840 × 7168
317addedAdd_52_ffn
Add
1 × 163840 × 7168
318addedRMSNorm_53_1
Rms Norm
7.2K1 × 163840 × 7168
319addedMLA_53
Mla
110M1 × 163840 × 7168
320addedAdd_53_attn
Add
1 × 163840 × 7168
321addedRMSNorm_53_2
Rms Norm
7.2K1 × 163840 × 7168
322addedMoE_53
Shared Expert Mo E
11B1 × 163840 × 7168
323addedAdd_53_ffn
Add
1 × 163840 × 7168
324addedRMSNorm_54_1
Rms Norm
7.2K1 × 163840 × 7168
325addedMLA_54
Mla
110M1 × 163840 × 7168
326addedAdd_54_attn
Add
1 × 163840 × 7168
327addedRMSNorm_54_2
Rms Norm
7.2K1 × 163840 × 7168
328addedMoE_54
Shared Expert Mo E
11B1 × 163840 × 7168
329addedAdd_54_ffn
Add
1 × 163840 × 7168
330addedRMSNorm_55_1
Rms Norm
7.2K1 × 163840 × 7168
331addedMLA_55
Mla
110M1 × 163840 × 7168
332addedAdd_55_attn
Add
1 × 163840 × 7168
333addedRMSNorm_55_2
Rms Norm
7.2K1 × 163840 × 7168
334addedMoE_55
Shared Expert Mo E
11B1 × 163840 × 7168
335addedAdd_55_ffn
Add
1 × 163840 × 7168
336addedRMSNorm_56_1
Rms Norm
7.2K1 × 163840 × 7168
337addedMLA_56
Mla
110M1 × 163840 × 7168
338addedAdd_56_attn
Add
1 × 163840 × 7168
339addedRMSNorm_56_2
Rms Norm
7.2K1 × 163840 × 7168
340addedMoE_56
Shared Expert Mo E
11B1 × 163840 × 7168
341addedAdd_56_ffn
Add
1 × 163840 × 7168
342addedRMSNorm_57_1
Rms Norm
7.2K1 × 163840 × 7168
343addedMLA_57
Mla
110M1 × 163840 × 7168
344addedAdd_57_attn
Add
1 × 163840 × 7168
345addedRMSNorm_57_2
Rms Norm
7.2K1 × 163840 × 7168
346addedMoE_57
Shared Expert Mo E
11B1 × 163840 × 7168
347addedAdd_57_ffn
Add
1 × 163840 × 7168
348addedRMSNorm_58_1
Rms Norm
7.2K1 × 163840 × 7168
349addedMLA_58
Mla
110M1 × 163840 × 7168
350addedAdd_58_attn
Add
1 × 163840 × 7168
351addedRMSNorm_58_2
Rms Norm
7.2K1 × 163840 × 7168
352addedMoE_58
Shared Expert Mo E
11B1 × 163840 × 7168
353addedAdd_58_ffn
Add
1 × 163840 × 7168
354addedRMSNorm_59_1
Rms Norm
7.2K1 × 163840 × 7168
355addedMLA_59
Mla
110M1 × 163840 × 7168
356addedAdd_59_attn
Add
1 × 163840 × 7168
357addedRMSNorm_59_2
Rms Norm
7.2K1 × 163840 × 7168
358addedMoE_59
Shared Expert Mo E
11B1 × 163840 × 7168
359addedAdd_59_ffn
Add
1 × 163840 × 7168
360addedRMSNorm_60_1
Rms Norm
7.2K1 × 163840 × 7168
361addedMLA_60
Mla
110M1 × 163840 × 7168
362addedAdd_60_attn
Add
1 × 163840 × 7168
363addedRMSNorm_60_2
Rms Norm
7.2K1 × 163840 × 7168
364addedMoE_60
Shared Expert Mo E
11B1 × 163840 × 7168
365addedAdd_60_ffn
Add
1 × 163840 × 7168
366addedRMSNorm_61_1
Rms Norm
7.2K1 × 163840 × 7168
367addedMLA_61
Mla
110M1 × 163840 × 7168
368addedAdd_61_attn
Add
1 × 163840 × 7168
369addedRMSNorm_61_2
Rms Norm
7.2K1 × 163840 × 7168
370addedMoE_61
Shared Expert Mo E
11B1 × 163840 × 7168
371samemoe_residual
Add
1 × 4096 × 4096Add_61_ffn
Add
1 × 163840 × 7168
372addedFinal_RMSNorm
Rms Norm
7.2K1 × 163840 × 7168
373addedLM_Head
Linear
927M1 × 163840 × 129280
374samehidden_out
Output
1 × 4096 × 4096Output
Output
1 × 163840 × 129280

What this is not

Take it further

Open either graph in the editor, change it, and check it again: Mixtral MoE Block · DeepSeek-V3

Compare any two models of your own, including anything on Hugging Face: the comparison tool.

Machine-readable: this page as markdown · the pair index · POST https://www.neurarch.com/api/v1/plan for a graph of your own.

Related comparisons

LLaMA-3 Block vs Mixtral MoE Block
A dense feed-forward against a mixture of experts, at the block level.
LLaMA-3 Block vs DeepSeek-V3
What a frontier open model adds to the block everyone started from.