Models / mimo_v2

MiMo-V2.5

Reconstructed from its own config.json with no weights read. 323K downloads on Hugging Face.

Our count against the checkpoint

The left number comes from the graph. The right one is the number of scalars in the published weight files. Nothing on this page was tuned to make them agree.

Derived from structure

316B

315,696,050,688 parameters

In the published checkpoint

311B

310,775,040,000 scalars · safetensors.total, read 2026-07-09

Delta

+1.58%

quantized This checkpoint is stored quantized (fp8). The tensor count in the file counts stored elements under a packing scheme, not logical parameters, so the two numbers below are not measuring the same thing in either direction.

What it costs to run

Cost is a roofline estimate on the priced GPU for 10 epochs at batch 32 over 50,000 samples (assumed; no dataset attached). GPU fit is fp32 weights plus gradients plus two Adam moments (16 bytes per parameter) with 1.3x headroom; activations are not included and grow with batch size.

Layers

300

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$24900539.68

Card	Memory
T4 (16GB)	weights + activations	does not fit
A100 (40GB)	weights + activations	does not fit
H100 (80GB)	weights + activations	does not fit

Structure

304 nodes. Output shapes are propagated from the input shape, batch dimension excluded.

	Layer	Type	Output shape
1	Input	Input	1 × 1048576
2	Embedding	Embedding	1 × 1048576 × 4096
3	RoPE	RoPE	1 × 1048576 × 4096
4	Vision input	Input	3 × 224 × 224
5	PatchEmbed	Patch Embed	196 × 1280
6	Patch_Position_Embedding	Learned Pos Embed	196 × 1280
7	Vision encoder (internals not in config)	Projection	196 × 4096
8	Vision tokens	Reshape	1 × 196 × 4096
9	Audio input	Input	1 × 128 × 1
10	Audio feature embed	Projection	1 × 128 × 4096
11	Audio encoder (internals not in config)	Projection	1 × 128 × 4096
12	Audio tokens	Reshape	1 × 256 × 4096
13	Multimodal fusion (concat tokens)	Concatenate	1 × 1049028 × 4096
14	RMSNorm_1_1	RMSNorm	1 × 1049028 × 4096
15	Attention_1	Grouped Query Attn	1 × 1049028 × 4096
16	Add_1_attn	Add	1 × 1049028 × 4096
17	RMSNorm_1_2	RMSNorm	1 × 1049028 × 4096
18	MoE_1	MoE Layer	1 × 1049028 × 4096
19	Add_1_ffn	Add	1 × 1049028 × 4096
20	RMSNorm_2_1	RMSNorm	1 × 1049028 × 4096
21	Attention_2	Grouped Query Attn	1 × 1049028 × 4096
22	Add_2_attn	Add	1 × 1049028 × 4096
23	RMSNorm_2_2	RMSNorm	1 × 1049028 × 4096
24	MoE_2	MoE Layer	1 × 1049028 × 4096
25	Add_2_ffn	Add	1 × 1049028 × 4096
26	RMSNorm_3_1	RMSNorm	1 × 1049028 × 4096
27	Attention_3	Grouped Query Attn	1 × 1049028 × 4096
28	Add_3_attn	Add	1 × 1049028 × 4096
29	RMSNorm_3_2	RMSNorm	1 × 1049028 × 4096
30	MoE_3	MoE Layer	1 × 1049028 × 4096
31	Add_3_ffn	Add	1 × 1049028 × 4096
32	RMSNorm_4_1	RMSNorm	1 × 1049028 × 4096
33	Attention_4	Grouped Query Attn	1 × 1049028 × 4096
34	Add_4_attn	Add	1 × 1049028 × 4096
35	RMSNorm_4_2	RMSNorm	1 × 1049028 × 4096
36	MoE_4	MoE Layer	1 × 1049028 × 4096
37	Add_4_ffn	Add	1 × 1049028 × 4096
38	RMSNorm_5_1	RMSNorm	1 × 1049028 × 4096
39	Attention_5	Grouped Query Attn	1 × 1049028 × 4096
40	Add_5_attn	Add	1 × 1049028 × 4096
41	RMSNorm_5_2	RMSNorm	1 × 1049028 × 4096
42	MoE_5	MoE Layer	1 × 1049028 × 4096
43	Add_5_ffn	Add	1 × 1049028 × 4096
44	RMSNorm_6_1	RMSNorm	1 × 1049028 × 4096
45	Attention_6	Grouped Query Attn	1 × 1049028 × 4096
46	Add_6_attn	Add	1 × 1049028 × 4096
47	RMSNorm_6_2	RMSNorm	1 × 1049028 × 4096
48	MoE_6	MoE Layer	1 × 1049028 × 4096
49	Add_6_ffn	Add	1 × 1049028 × 4096
50	RMSNorm_7_1	RMSNorm	1 × 1049028 × 4096
51	Attention_7	Grouped Query Attn	1 × 1049028 × 4096
52	Add_7_attn	Add	1 × 1049028 × 4096
53	RMSNorm_7_2	RMSNorm	1 × 1049028 × 4096
54	MoE_7	MoE Layer	1 × 1049028 × 4096
55	Add_7_ffn	Add	1 × 1049028 × 4096
56	RMSNorm_8_1	RMSNorm	1 × 1049028 × 4096
57	Attention_8	Grouped Query Attn	1 × 1049028 × 4096
58	Add_8_attn	Add	1 × 1049028 × 4096
59	RMSNorm_8_2	RMSNorm	1 × 1049028 × 4096
60	MoE_8	MoE Layer	1 × 1049028 × 4096
61	Add_8_ffn	Add	1 × 1049028 × 4096
62	RMSNorm_9_1	RMSNorm	1 × 1049028 × 4096
63	Attention_9	Grouped Query Attn	1 × 1049028 × 4096
64	Add_9_attn	Add	1 × 1049028 × 4096
65	RMSNorm_9_2	RMSNorm	1 × 1049028 × 4096
66	MoE_9	MoE Layer	1 × 1049028 × 4096
67	Add_9_ffn	Add	1 × 1049028 × 4096
68	RMSNorm_10_1	RMSNorm	1 × 1049028 × 4096
69	Attention_10	Grouped Query Attn	1 × 1049028 × 4096
70	Add_10_attn	Add	1 × 1049028 × 4096
71	RMSNorm_10_2	RMSNorm	1 × 1049028 × 4096
72	MoE_10	MoE Layer	1 × 1049028 × 4096
73	Add_10_ffn	Add	1 × 1049028 × 4096
74	RMSNorm_11_1	RMSNorm	1 × 1049028 × 4096
75	Attention_11	Grouped Query Attn	1 × 1049028 × 4096
76	Add_11_attn	Add	1 × 1049028 × 4096
77	RMSNorm_11_2	RMSNorm	1 × 1049028 × 4096
78	MoE_11	MoE Layer	1 × 1049028 × 4096
79	Add_11_ffn	Add	1 × 1049028 × 4096
80	RMSNorm_12_1	RMSNorm	1 × 1049028 × 4096
81	Attention_12	Grouped Query Attn	1 × 1049028 × 4096
82	Add_12_attn	Add	1 × 1049028 × 4096
83	RMSNorm_12_2	RMSNorm	1 × 1049028 × 4096
84	MoE_12	MoE Layer	1 × 1049028 × 4096
85	Add_12_ffn	Add	1 × 1049028 × 4096
86	RMSNorm_13_1	RMSNorm	1 × 1049028 × 4096
87	Attention_13	Grouped Query Attn	1 × 1049028 × 4096
88	Add_13_attn	Add	1 × 1049028 × 4096
89	RMSNorm_13_2	RMSNorm	1 × 1049028 × 4096
90	MoE_13	MoE Layer	1 × 1049028 × 4096
91	Add_13_ffn	Add	1 × 1049028 × 4096
92	RMSNorm_14_1	RMSNorm	1 × 1049028 × 4096
93	Attention_14	Grouped Query Attn	1 × 1049028 × 4096
94	Add_14_attn	Add	1 × 1049028 × 4096
95	RMSNorm_14_2	RMSNorm	1 × 1049028 × 4096
96	MoE_14	MoE Layer	1 × 1049028 × 4096
97	Add_14_ffn	Add	1 × 1049028 × 4096
98	RMSNorm_15_1	RMSNorm	1 × 1049028 × 4096
99	Attention_15	Grouped Query Attn	1 × 1049028 × 4096
100	Add_15_attn	Add	1 × 1049028 × 4096
101	RMSNorm_15_2	RMSNorm	1 × 1049028 × 4096
102	MoE_15	MoE Layer	1 × 1049028 × 4096
103	Add_15_ffn	Add	1 × 1049028 × 4096
104	RMSNorm_16_1	RMSNorm	1 × 1049028 × 4096
105	Attention_16	Grouped Query Attn	1 × 1049028 × 4096
106	Add_16_attn	Add	1 × 1049028 × 4096
107	RMSNorm_16_2	RMSNorm	1 × 1049028 × 4096
108	MoE_16	MoE Layer	1 × 1049028 × 4096
109	Add_16_ffn	Add	1 × 1049028 × 4096
110	RMSNorm_17_1	RMSNorm	1 × 1049028 × 4096
111	Attention_17	Grouped Query Attn	1 × 1049028 × 4096
112	Add_17_attn	Add	1 × 1049028 × 4096
113	RMSNorm_17_2	RMSNorm	1 × 1049028 × 4096
114	MoE_17	MoE Layer	1 × 1049028 × 4096
115	Add_17_ffn	Add	1 × 1049028 × 4096
116	RMSNorm_18_1	RMSNorm	1 × 1049028 × 4096
117	Attention_18	Grouped Query Attn	1 × 1049028 × 4096
118	Add_18_attn	Add	1 × 1049028 × 4096
119	RMSNorm_18_2	RMSNorm	1 × 1049028 × 4096
120	MoE_18	MoE Layer	1 × 1049028 × 4096
121	Add_18_ffn	Add	1 × 1049028 × 4096
122	RMSNorm_19_1	RMSNorm	1 × 1049028 × 4096
123	Attention_19	Grouped Query Attn	1 × 1049028 × 4096
124	Add_19_attn	Add	1 × 1049028 × 4096
125	RMSNorm_19_2	RMSNorm	1 × 1049028 × 4096
126	MoE_19	MoE Layer	1 × 1049028 × 4096
127	Add_19_ffn	Add	1 × 1049028 × 4096
128	RMSNorm_20_1	RMSNorm	1 × 1049028 × 4096
129	Attention_20	Grouped Query Attn	1 × 1049028 × 4096
130	Add_20_attn	Add	1 × 1049028 × 4096
131	RMSNorm_20_2	RMSNorm	1 × 1049028 × 4096
132	MoE_20	MoE Layer	1 × 1049028 × 4096
133	Add_20_ffn	Add	1 × 1049028 × 4096
134	RMSNorm_21_1	RMSNorm	1 × 1049028 × 4096
135	Attention_21	Grouped Query Attn	1 × 1049028 × 4096
136	Add_21_attn	Add	1 × 1049028 × 4096
137	RMSNorm_21_2	RMSNorm	1 × 1049028 × 4096
138	MoE_21	MoE Layer	1 × 1049028 × 4096
139	Add_21_ffn	Add	1 × 1049028 × 4096
140	RMSNorm_22_1	RMSNorm	1 × 1049028 × 4096
141	Attention_22	Grouped Query Attn	1 × 1049028 × 4096
142	Add_22_attn	Add	1 × 1049028 × 4096
143	RMSNorm_22_2	RMSNorm	1 × 1049028 × 4096
144	MoE_22	MoE Layer	1 × 1049028 × 4096
145	Add_22_ffn	Add	1 × 1049028 × 4096
146	RMSNorm_23_1	RMSNorm	1 × 1049028 × 4096
147	Attention_23	Grouped Query Attn	1 × 1049028 × 4096
148	Add_23_attn	Add	1 × 1049028 × 4096
149	RMSNorm_23_2	RMSNorm	1 × 1049028 × 4096
150	MoE_23	MoE Layer	1 × 1049028 × 4096
151	Add_23_ffn	Add	1 × 1049028 × 4096
152	RMSNorm_24_1	RMSNorm	1 × 1049028 × 4096
153	Attention_24	Grouped Query Attn	1 × 1049028 × 4096
154	Add_24_attn	Add	1 × 1049028 × 4096
155	RMSNorm_24_2	RMSNorm	1 × 1049028 × 4096
156	MoE_24	MoE Layer	1 × 1049028 × 4096
157	Add_24_ffn	Add	1 × 1049028 × 4096
158	RMSNorm_25_1	RMSNorm	1 × 1049028 × 4096
159	Attention_25	Grouped Query Attn	1 × 1049028 × 4096
160	Add_25_attn	Add	1 × 1049028 × 4096
161	RMSNorm_25_2	RMSNorm	1 × 1049028 × 4096
162	MoE_25	MoE Layer	1 × 1049028 × 4096
163	Add_25_ffn	Add	1 × 1049028 × 4096
164	RMSNorm_26_1	RMSNorm	1 × 1049028 × 4096
165	Attention_26	Grouped Query Attn	1 × 1049028 × 4096
166	Add_26_attn	Add	1 × 1049028 × 4096
167	RMSNorm_26_2	RMSNorm	1 × 1049028 × 4096
168	MoE_26	MoE Layer	1 × 1049028 × 4096
169	Add_26_ffn	Add	1 × 1049028 × 4096
170	RMSNorm_27_1	RMSNorm	1 × 1049028 × 4096
171	Attention_27	Grouped Query Attn	1 × 1049028 × 4096
172	Add_27_attn	Add	1 × 1049028 × 4096
173	RMSNorm_27_2	RMSNorm	1 × 1049028 × 4096
174	MoE_27	MoE Layer	1 × 1049028 × 4096
175	Add_27_ffn	Add	1 × 1049028 × 4096
176	RMSNorm_28_1	RMSNorm	1 × 1049028 × 4096
177	Attention_28	Grouped Query Attn	1 × 1049028 × 4096
178	Add_28_attn	Add	1 × 1049028 × 4096
179	RMSNorm_28_2	RMSNorm	1 × 1049028 × 4096
180	MoE_28	MoE Layer	1 × 1049028 × 4096
181	Add_28_ffn	Add	1 × 1049028 × 4096
182	RMSNorm_29_1	RMSNorm	1 × 1049028 × 4096
183	Attention_29	Grouped Query Attn	1 × 1049028 × 4096
184	Add_29_attn	Add	1 × 1049028 × 4096
185	RMSNorm_29_2	RMSNorm	1 × 1049028 × 4096
186	MoE_29	MoE Layer	1 × 1049028 × 4096
187	Add_29_ffn	Add	1 × 1049028 × 4096
188	RMSNorm_30_1	RMSNorm	1 × 1049028 × 4096
189	Attention_30	Grouped Query Attn	1 × 1049028 × 4096
190	Add_30_attn	Add	1 × 1049028 × 4096
191	RMSNorm_30_2	RMSNorm	1 × 1049028 × 4096
192	MoE_30	MoE Layer	1 × 1049028 × 4096
193	Add_30_ffn	Add	1 × 1049028 × 4096
194	RMSNorm_31_1	RMSNorm	1 × 1049028 × 4096
195	Attention_31	Grouped Query Attn	1 × 1049028 × 4096
196	Add_31_attn	Add	1 × 1049028 × 4096
197	RMSNorm_31_2	RMSNorm	1 × 1049028 × 4096
198	MoE_31	MoE Layer	1 × 1049028 × 4096
199	Add_31_ffn	Add	1 × 1049028 × 4096
200	RMSNorm_32_1	RMSNorm	1 × 1049028 × 4096
201	Attention_32	Grouped Query Attn	1 × 1049028 × 4096
202	Add_32_attn	Add	1 × 1049028 × 4096
203	RMSNorm_32_2	RMSNorm	1 × 1049028 × 4096
204	MoE_32	MoE Layer	1 × 1049028 × 4096
205	Add_32_ffn	Add	1 × 1049028 × 4096
206	RMSNorm_33_1	RMSNorm	1 × 1049028 × 4096
207	Attention_33	Grouped Query Attn	1 × 1049028 × 4096
208	Add_33_attn	Add	1 × 1049028 × 4096
209	RMSNorm_33_2	RMSNorm	1 × 1049028 × 4096
210	MoE_33	MoE Layer	1 × 1049028 × 4096
211	Add_33_ffn	Add	1 × 1049028 × 4096
212	RMSNorm_34_1	RMSNorm	1 × 1049028 × 4096
213	Attention_34	Grouped Query Attn	1 × 1049028 × 4096
214	Add_34_attn	Add	1 × 1049028 × 4096
215	RMSNorm_34_2	RMSNorm	1 × 1049028 × 4096
216	MoE_34	MoE Layer	1 × 1049028 × 4096
217	Add_34_ffn	Add	1 × 1049028 × 4096
218	RMSNorm_35_1	RMSNorm	1 × 1049028 × 4096
219	Attention_35	Grouped Query Attn	1 × 1049028 × 4096
220	Add_35_attn	Add	1 × 1049028 × 4096
221	RMSNorm_35_2	RMSNorm	1 × 1049028 × 4096
222	MoE_35	MoE Layer	1 × 1049028 × 4096
223	Add_35_ffn	Add	1 × 1049028 × 4096
224	RMSNorm_36_1	RMSNorm	1 × 1049028 × 4096
225	Attention_36	Grouped Query Attn	1 × 1049028 × 4096
226	Add_36_attn	Add	1 × 1049028 × 4096
227	RMSNorm_36_2	RMSNorm	1 × 1049028 × 4096
228	MoE_36	MoE Layer	1 × 1049028 × 4096
229	Add_36_ffn	Add	1 × 1049028 × 4096
230	RMSNorm_37_1	RMSNorm	1 × 1049028 × 4096
231	Attention_37	Grouped Query Attn	1 × 1049028 × 4096
232	Add_37_attn	Add	1 × 1049028 × 4096
233	RMSNorm_37_2	RMSNorm	1 × 1049028 × 4096
234	MoE_37	MoE Layer	1 × 1049028 × 4096
235	Add_37_ffn	Add	1 × 1049028 × 4096
236	RMSNorm_38_1	RMSNorm	1 × 1049028 × 4096
237	Attention_38	Grouped Query Attn	1 × 1049028 × 4096
238	Add_38_attn	Add	1 × 1049028 × 4096
239	RMSNorm_38_2	RMSNorm	1 × 1049028 × 4096
240	MoE_38	MoE Layer	1 × 1049028 × 4096
241	Add_38_ffn	Add	1 × 1049028 × 4096
242	RMSNorm_39_1	RMSNorm	1 × 1049028 × 4096
243	Attention_39	Grouped Query Attn	1 × 1049028 × 4096
244	Add_39_attn	Add	1 × 1049028 × 4096
245	RMSNorm_39_2	RMSNorm	1 × 1049028 × 4096
246	MoE_39	MoE Layer	1 × 1049028 × 4096
247	Add_39_ffn	Add	1 × 1049028 × 4096
248	RMSNorm_40_1	RMSNorm	1 × 1049028 × 4096
249	Attention_40	Grouped Query Attn	1 × 1049028 × 4096
250	Add_40_attn	Add	1 × 1049028 × 4096
251	RMSNorm_40_2	RMSNorm	1 × 1049028 × 4096
252	MoE_40	MoE Layer	1 × 1049028 × 4096
253	Add_40_ffn	Add	1 × 1049028 × 4096
254	RMSNorm_41_1	RMSNorm	1 × 1049028 × 4096
255	Attention_41	Grouped Query Attn	1 × 1049028 × 4096
256	Add_41_attn	Add	1 × 1049028 × 4096
257	RMSNorm_41_2	RMSNorm	1 × 1049028 × 4096
258	MoE_41	MoE Layer	1 × 1049028 × 4096
259	Add_41_ffn	Add	1 × 1049028 × 4096
260	RMSNorm_42_1	RMSNorm	1 × 1049028 × 4096
261	Attention_42	Grouped Query Attn	1 × 1049028 × 4096
262	Add_42_attn	Add	1 × 1049028 × 4096
263	RMSNorm_42_2	RMSNorm	1 × 1049028 × 4096
264	MoE_42	MoE Layer	1 × 1049028 × 4096
265	Add_42_ffn	Add	1 × 1049028 × 4096
266	RMSNorm_43_1	RMSNorm	1 × 1049028 × 4096
267	Attention_43	Grouped Query Attn	1 × 1049028 × 4096
268	Add_43_attn	Add	1 × 1049028 × 4096
269	RMSNorm_43_2	RMSNorm	1 × 1049028 × 4096
270	MoE_43	MoE Layer	1 × 1049028 × 4096
271	Add_43_ffn	Add	1 × 1049028 × 4096
272	RMSNorm_44_1	RMSNorm	1 × 1049028 × 4096
273	Attention_44	Grouped Query Attn	1 × 1049028 × 4096
274	Add_44_attn	Add	1 × 1049028 × 4096
275	RMSNorm_44_2	RMSNorm	1 × 1049028 × 4096
276	MoE_44	MoE Layer	1 × 1049028 × 4096
277	Add_44_ffn	Add	1 × 1049028 × 4096
278	RMSNorm_45_1	RMSNorm	1 × 1049028 × 4096
279	Attention_45	Grouped Query Attn	1 × 1049028 × 4096
280	Add_45_attn	Add	1 × 1049028 × 4096
281	RMSNorm_45_2	RMSNorm	1 × 1049028 × 4096
282	MoE_45	MoE Layer	1 × 1049028 × 4096
283	Add_45_ffn	Add	1 × 1049028 × 4096
284	RMSNorm_46_1	RMSNorm	1 × 1049028 × 4096
285	Attention_46	Grouped Query Attn	1 × 1049028 × 4096
286	Add_46_attn	Add	1 × 1049028 × 4096
287	RMSNorm_46_2	RMSNorm	1 × 1049028 × 4096
288	MoE_46	MoE Layer	1 × 1049028 × 4096
289	Add_46_ffn	Add	1 × 1049028 × 4096
290	RMSNorm_47_1	RMSNorm	1 × 1049028 × 4096
291	Attention_47	Grouped Query Attn	1 × 1049028 × 4096
292	Add_47_attn	Add	1 × 1049028 × 4096
293	RMSNorm_47_2	RMSNorm	1 × 1049028 × 4096
294	MoE_47	MoE Layer	1 × 1049028 × 4096
295	Add_47_ffn	Add	1 × 1049028 × 4096
296	RMSNorm_48_1	RMSNorm	1 × 1049028 × 4096
297	Attention_48	Grouped Query Attn	1 × 1049028 × 4096
298	Add_48_attn	Add	1 × 1049028 × 4096
299	RMSNorm_48_2	RMSNorm	1 × 1049028 × 4096
300	MoE_48	MoE Layer	1 × 1049028 × 4096
301	Add_48_ffn	Add	1 × 1049028 × 4096
302	Final_RMSNorm	RMSNorm	1 × 1049028 × 4096
303	LM_Head	Linear	1 × 1049028 × 152576
304	Output	Output	1 × 1049028 × 152576

What the verifier says

warn"RoPE" receives input but its output is not connected. This layer will be unreachable in the forward pass. Fix: Connect the output forward, or add an Output node if this is the final layer.
dead-end

infoMoE layers require an auxiliary router z-loss + load-balance loss during training to prevent expert collapse. This is not visible in the architecture diagram but must be in the training loop. Applies to all 48: MoE_1, MoE_2, MoE_3, MoE_4, MoE_5, MoE_6, MoE_7, MoE_8, MoE_9, MoE_10, MoE_11, MoE_12, MoE_13, MoE_14, MoE_15, MoE_16, MoE_17, MoE_18, MoE_19, MoE_20, MoE_21, MoE_22, MoE_23, MoE_24, MoE_25, MoE_26, MoE_27, MoE_28, MoE_29, MoE_30, MoE_31, MoE_32, MoE_33, MoE_34, MoE_35, MoE_36, MoE_37, MoE_38, MoE_39, MoE_40, MoE_41, MoE_42, MoE_43, MoE_44, MoE_45, MoE_46, MoE_47, MoE_48. Fix: Add a note on these layers. Typical aux_loss coefficient: 1e-2 (Mixtral/Switch Transformer).
moe-no-aux-loss

infoAt 48 stacked attention layers, residual-branch outputs add up; unscaled init lets activation variance grow with depth. GPT-2/LLaMA-family models scale the residual projections by depth (N(0, 0.02 / √(2L))). Fix: Scale residual output projections by depth: nn.init.normal_(w, std=0.02 / math.sqrt(2 * n_layers))
deep-attention-default-init

Do this to your own model

Same numbers, on a model in your repo, in one command. No account.

pip install neurarch-trace
neurarch-trace XiaomiMiMo/MiMo-V2.5 --plan --share

Paste a model instead Markdown twin Index JSON POST /api/v1/plan