Models / minicpmo

MiniCPM-o-4_5

Reconstructed from its own config.json with no weights read. 753K 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

8.92B

8,918,330,932 parameters

In the published checkpoint

9.37B

9,371,787,666 scalars · safetensors.total, read 2026-08-18

Delta

-4.84%

custom-code This repository ships its own modeling code (`auto_map`, e.g. `configuration_minicpmo.py`), so `config.json` names a class in the repo rather than an architecture `transformers` defines. The graph below is what those config keys mean under `transformers` semantics, which is not necessarily what the repo's own file builds. A gap here is a statement about what we read, not about the checkpoint.

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

432

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$24904.09

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

Structure

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

	Layer	Type	Output shape
1	Input	Input	1 × 40960
2	Embedding	Embedding	1 × 40960 × 4096
3	RoPE	RoPE	1 × 40960 × 4096
4	Vision input	Input	3 × 980 × 980
5	PatchEmbed	Patch Embed	4900 × 1152
6	Patch_Position_Embedding	Learned Pos Embed	4900 × 1152
7	Vision_LN_1	LayerNorm	4900 × 1152
8	Vision_Attn_1	Multi-Head Attention	4900 × 1152
9	Vision_Add_1	Add	4900 × 1152
10	Vision_FFN_1	Feed Forward	4900 × 1152
11	Vision_LN_2	LayerNorm	4900 × 1152
12	Vision_Attn_2	Multi-Head Attention	4900 × 1152
13	Vision_Add_2	Add	4900 × 1152
14	Vision_FFN_2	Feed Forward	4900 × 1152
15	Vision_LN_3	LayerNorm	4900 × 1152
16	Vision_Attn_3	Multi-Head Attention	4900 × 1152
17	Vision_Add_3	Add	4900 × 1152
18	Vision_FFN_3	Feed Forward	4900 × 1152
19	Vision_LN_4	LayerNorm	4900 × 1152
20	Vision_Attn_4	Multi-Head Attention	4900 × 1152
21	Vision_Add_4	Add	4900 × 1152
22	Vision_FFN_4	Feed Forward	4900 × 1152
23	Vision_LN_5	LayerNorm	4900 × 1152
24	Vision_Attn_5	Multi-Head Attention	4900 × 1152
25	Vision_Add_5	Add	4900 × 1152
26	Vision_FFN_5	Feed Forward	4900 × 1152
27	Vision_LN_6	LayerNorm	4900 × 1152
28	Vision_Attn_6	Multi-Head Attention	4900 × 1152
29	Vision_Add_6	Add	4900 × 1152
30	Vision_FFN_6	Feed Forward	4900 × 1152
31	Vision_LN_7	LayerNorm	4900 × 1152
32	Vision_Attn_7	Multi-Head Attention	4900 × 1152
33	Vision_Add_7	Add	4900 × 1152
34	Vision_FFN_7	Feed Forward	4900 × 1152
35	Vision_LN_8	LayerNorm	4900 × 1152
36	Vision_Attn_8	Multi-Head Attention	4900 × 1152
37	Vision_Add_8	Add	4900 × 1152
38	Vision_FFN_8	Feed Forward	4900 × 1152
39	Vision_LN_9	LayerNorm	4900 × 1152
40	Vision_Attn_9	Multi-Head Attention	4900 × 1152
41	Vision_Add_9	Add	4900 × 1152
42	Vision_FFN_9	Feed Forward	4900 × 1152
43	Vision_LN_10	LayerNorm	4900 × 1152
44	Vision_Attn_10	Multi-Head Attention	4900 × 1152
45	Vision_Add_10	Add	4900 × 1152
46	Vision_FFN_10	Feed Forward	4900 × 1152
47	Vision_LN_11	LayerNorm	4900 × 1152
48	Vision_Attn_11	Multi-Head Attention	4900 × 1152
49	Vision_Add_11	Add	4900 × 1152
50	Vision_FFN_11	Feed Forward	4900 × 1152
51	Vision_LN_12	LayerNorm	4900 × 1152
52	Vision_Attn_12	Multi-Head Attention	4900 × 1152
53	Vision_Add_12	Add	4900 × 1152
54	Vision_FFN_12	Feed Forward	4900 × 1152
55	Vision_LN_13	LayerNorm	4900 × 1152
56	Vision_Attn_13	Multi-Head Attention	4900 × 1152
57	Vision_Add_13	Add	4900 × 1152
58	Vision_FFN_13	Feed Forward	4900 × 1152
59	Vision_LN_14	LayerNorm	4900 × 1152
60	Vision_Attn_14	Multi-Head Attention	4900 × 1152
61	Vision_Add_14	Add	4900 × 1152
62	Vision_FFN_14	Feed Forward	4900 × 1152
63	Vision_LN_15	LayerNorm	4900 × 1152
64	Vision_Attn_15	Multi-Head Attention	4900 × 1152
65	Vision_Add_15	Add	4900 × 1152
66	Vision_FFN_15	Feed Forward	4900 × 1152
67	Vision_LN_16	LayerNorm	4900 × 1152
68	Vision_Attn_16	Multi-Head Attention	4900 × 1152
69	Vision_Add_16	Add	4900 × 1152
70	Vision_FFN_16	Feed Forward	4900 × 1152
71	Vision_LN_17	LayerNorm	4900 × 1152
72	Vision_Attn_17	Multi-Head Attention	4900 × 1152
73	Vision_Add_17	Add	4900 × 1152
74	Vision_FFN_17	Feed Forward	4900 × 1152
75	Vision_LN_18	LayerNorm	4900 × 1152
76	Vision_Attn_18	Multi-Head Attention	4900 × 1152
77	Vision_Add_18	Add	4900 × 1152
78	Vision_FFN_18	Feed Forward	4900 × 1152
79	Vision_LN_19	LayerNorm	4900 × 1152
80	Vision_Attn_19	Multi-Head Attention	4900 × 1152
81	Vision_Add_19	Add	4900 × 1152
82	Vision_FFN_19	Feed Forward	4900 × 1152
83	Vision_LN_20	LayerNorm	4900 × 1152
84	Vision_Attn_20	Multi-Head Attention	4900 × 1152
85	Vision_Add_20	Add	4900 × 1152
86	Vision_FFN_20	Feed Forward	4900 × 1152
87	Vision_LN_21	LayerNorm	4900 × 1152
88	Vision_Attn_21	Multi-Head Attention	4900 × 1152
89	Vision_Add_21	Add	4900 × 1152
90	Vision_FFN_21	Feed Forward	4900 × 1152
91	Vision_LN_22	LayerNorm	4900 × 1152
92	Vision_Attn_22	Multi-Head Attention	4900 × 1152
93	Vision_Add_22	Add	4900 × 1152
94	Vision_FFN_22	Feed Forward	4900 × 1152
95	Vision_LN_23	LayerNorm	4900 × 1152
96	Vision_Attn_23	Multi-Head Attention	4900 × 1152
97	Vision_Add_23	Add	4900 × 1152
98	Vision_FFN_23	Feed Forward	4900 × 1152
99	Vision_LN_24	LayerNorm	4900 × 1152
100	Vision_Attn_24	Multi-Head Attention	4900 × 1152
101	Vision_Add_24	Add	4900 × 1152
102	Vision_FFN_24	Feed Forward	4900 × 1152
103	Vision_LN_25	LayerNorm	4900 × 1152
104	Vision_Attn_25	Multi-Head Attention	4900 × 1152
105	Vision_Add_25	Add	4900 × 1152
106	Vision_FFN_25	Feed Forward	4900 × 1152
107	Vision_LN_26	LayerNorm	4900 × 1152
108	Vision_Attn_26	Multi-Head Attention	4900 × 1152
109	Vision_Add_26	Add	4900 × 1152
110	Vision_FFN_26	Feed Forward	4900 × 1152
111	Vision_LN_27	LayerNorm	4900 × 1152
112	Vision_Attn_27	Multi-Head Attention	4900 × 1152
113	Vision_Add_27	Add	4900 × 1152
114	Vision_FFN_27	Feed Forward	4900 × 1152
115	Vision projector	Projection	4900 × 4096
116	Vision tokens	Reshape	1 × 4900 × 4096
117	Audio input	Input	1 × 80 × 1
118	Audio feature embed	Projection	1 × 80 × 1024
119	Audio_LN_1	LayerNorm	1 × 80 × 1024
120	Audio_Attn_1	Multi-Head Attention	1 × 80 × 1024
121	Audio_Add_1	Add	1 × 80 × 1024
122	Audio_FFN_1	Feed Forward	1 × 80 × 1024
123	Audio_LN_2	LayerNorm	1 × 80 × 1024
124	Audio_Attn_2	Multi-Head Attention	1 × 80 × 1024
125	Audio_Add_2	Add	1 × 80 × 1024
126	Audio_FFN_2	Feed Forward	1 × 80 × 1024
127	Audio_LN_3	LayerNorm	1 × 80 × 1024
128	Audio_Attn_3	Multi-Head Attention	1 × 80 × 1024
129	Audio_Add_3	Add	1 × 80 × 1024
130	Audio_FFN_3	Feed Forward	1 × 80 × 1024
131	Audio_LN_4	LayerNorm	1 × 80 × 1024
132	Audio_Attn_4	Multi-Head Attention	1 × 80 × 1024
133	Audio_Add_4	Add	1 × 80 × 1024
134	Audio_FFN_4	Feed Forward	1 × 80 × 1024
135	Audio_LN_5	LayerNorm	1 × 80 × 1024
136	Audio_Attn_5	Multi-Head Attention	1 × 80 × 1024
137	Audio_Add_5	Add	1 × 80 × 1024
138	Audio_FFN_5	Feed Forward	1 × 80 × 1024
139	Audio_LN_6	LayerNorm	1 × 80 × 1024
140	Audio_Attn_6	Multi-Head Attention	1 × 80 × 1024
141	Audio_Add_6	Add	1 × 80 × 1024
142	Audio_FFN_6	Feed Forward	1 × 80 × 1024
143	Audio_LN_7	LayerNorm	1 × 80 × 1024
144	Audio_Attn_7	Multi-Head Attention	1 × 80 × 1024
145	Audio_Add_7	Add	1 × 80 × 1024
146	Audio_FFN_7	Feed Forward	1 × 80 × 1024
147	Audio_LN_8	LayerNorm	1 × 80 × 1024
148	Audio_Attn_8	Multi-Head Attention	1 × 80 × 1024
149	Audio_Add_8	Add	1 × 80 × 1024
150	Audio_FFN_8	Feed Forward	1 × 80 × 1024
151	Audio_LN_9	LayerNorm	1 × 80 × 1024
152	Audio_Attn_9	Multi-Head Attention	1 × 80 × 1024
153	Audio_Add_9	Add	1 × 80 × 1024
154	Audio_FFN_9	Feed Forward	1 × 80 × 1024
155	Audio_LN_10	LayerNorm	1 × 80 × 1024
156	Audio_Attn_10	Multi-Head Attention	1 × 80 × 1024
157	Audio_Add_10	Add	1 × 80 × 1024
158	Audio_FFN_10	Feed Forward	1 × 80 × 1024
159	Audio_LN_11	LayerNorm	1 × 80 × 1024
160	Audio_Attn_11	Multi-Head Attention	1 × 80 × 1024
161	Audio_Add_11	Add	1 × 80 × 1024
162	Audio_FFN_11	Feed Forward	1 × 80 × 1024
163	Audio_LN_12	LayerNorm	1 × 80 × 1024
164	Audio_Attn_12	Multi-Head Attention	1 × 80 × 1024
165	Audio_Add_12	Add	1 × 80 × 1024
166	Audio_FFN_12	Feed Forward	1 × 80 × 1024
167	Audio_LN_13	LayerNorm	1 × 80 × 1024
168	Audio_Attn_13	Multi-Head Attention	1 × 80 × 1024
169	Audio_Add_13	Add	1 × 80 × 1024
170	Audio_FFN_13	Feed Forward	1 × 80 × 1024
171	Audio_LN_14	LayerNorm	1 × 80 × 1024
172	Audio_Attn_14	Multi-Head Attention	1 × 80 × 1024
173	Audio_Add_14	Add	1 × 80 × 1024
174	Audio_FFN_14	Feed Forward	1 × 80 × 1024
175	Audio_LN_15	LayerNorm	1 × 80 × 1024
176	Audio_Attn_15	Multi-Head Attention	1 × 80 × 1024
177	Audio_Add_15	Add	1 × 80 × 1024
178	Audio_FFN_15	Feed Forward	1 × 80 × 1024
179	Audio_LN_16	LayerNorm	1 × 80 × 1024
180	Audio_Attn_16	Multi-Head Attention	1 × 80 × 1024
181	Audio_Add_16	Add	1 × 80 × 1024
182	Audio_FFN_16	Feed Forward	1 × 80 × 1024
183	Audio_LN_17	LayerNorm	1 × 80 × 1024
184	Audio_Attn_17	Multi-Head Attention	1 × 80 × 1024
185	Audio_Add_17	Add	1 × 80 × 1024
186	Audio_FFN_17	Feed Forward	1 × 80 × 1024
187	Audio_LN_18	LayerNorm	1 × 80 × 1024
188	Audio_Attn_18	Multi-Head Attention	1 × 80 × 1024
189	Audio_Add_18	Add	1 × 80 × 1024
190	Audio_FFN_18	Feed Forward	1 × 80 × 1024
191	Audio_LN_19	LayerNorm	1 × 80 × 1024
192	Audio_Attn_19	Multi-Head Attention	1 × 80 × 1024
193	Audio_Add_19	Add	1 × 80 × 1024
194	Audio_FFN_19	Feed Forward	1 × 80 × 1024
195	Audio_LN_20	LayerNorm	1 × 80 × 1024
196	Audio_Attn_20	Multi-Head Attention	1 × 80 × 1024
197	Audio_Add_20	Add	1 × 80 × 1024
198	Audio_FFN_20	Feed Forward	1 × 80 × 1024
199	Audio_LN_21	LayerNorm	1 × 80 × 1024
200	Audio_Attn_21	Multi-Head Attention	1 × 80 × 1024
201	Audio_Add_21	Add	1 × 80 × 1024
202	Audio_FFN_21	Feed Forward	1 × 80 × 1024
203	Audio_LN_22	LayerNorm	1 × 80 × 1024
204	Audio_Attn_22	Multi-Head Attention	1 × 80 × 1024
205	Audio_Add_22	Add	1 × 80 × 1024
206	Audio_FFN_22	Feed Forward	1 × 80 × 1024
207	Audio_LN_23	LayerNorm	1 × 80 × 1024
208	Audio_Attn_23	Multi-Head Attention	1 × 80 × 1024
209	Audio_Add_23	Add	1 × 80 × 1024
210	Audio_FFN_23	Feed Forward	1 × 80 × 1024
211	Audio_LN_24	LayerNorm	1 × 80 × 1024
212	Audio_Attn_24	Multi-Head Attention	1 × 80 × 1024
213	Audio_Add_24	Add	1 × 80 × 1024
214	Audio_FFN_24	Feed Forward	1 × 80 × 1024
215	Audio projector	Projection	1 × 80 × 4096
216	Audio tokens	Reshape	1 × 256 × 4096
217	Multimodal fusion (concat tokens)	Concatenate	1 × 46116 × 4096
218	RMSNorm_1_1	RMSNorm	1 × 46116 × 4096
219	Attention_1	Grouped Query Attn	1 × 46116 × 4096
220	Add_1_attn	Add	1 × 46116 × 4096
221	RMSNorm_1_2	RMSNorm	1 × 46116 × 4096
222	FFN_1	SwiGLU	1 × 46116 × 4096
223	Add_1_ffn	Add	1 × 46116 × 4096
224	RMSNorm_2_1	RMSNorm	1 × 46116 × 4096
225	Attention_2	Grouped Query Attn	1 × 46116 × 4096
226	Add_2_attn	Add	1 × 46116 × 4096
227	RMSNorm_2_2	RMSNorm	1 × 46116 × 4096
228	FFN_2	SwiGLU	1 × 46116 × 4096
229	Add_2_ffn	Add	1 × 46116 × 4096
230	RMSNorm_3_1	RMSNorm	1 × 46116 × 4096
231	Attention_3	Grouped Query Attn	1 × 46116 × 4096
232	Add_3_attn	Add	1 × 46116 × 4096
233	RMSNorm_3_2	RMSNorm	1 × 46116 × 4096
234	FFN_3	SwiGLU	1 × 46116 × 4096
235	Add_3_ffn	Add	1 × 46116 × 4096
236	RMSNorm_4_1	RMSNorm	1 × 46116 × 4096
237	Attention_4	Grouped Query Attn	1 × 46116 × 4096
238	Add_4_attn	Add	1 × 46116 × 4096
239	RMSNorm_4_2	RMSNorm	1 × 46116 × 4096
240	FFN_4	SwiGLU	1 × 46116 × 4096
241	Add_4_ffn	Add	1 × 46116 × 4096
242	RMSNorm_5_1	RMSNorm	1 × 46116 × 4096
243	Attention_5	Grouped Query Attn	1 × 46116 × 4096
244	Add_5_attn	Add	1 × 46116 × 4096
245	RMSNorm_5_2	RMSNorm	1 × 46116 × 4096
246	FFN_5	SwiGLU	1 × 46116 × 4096
247	Add_5_ffn	Add	1 × 46116 × 4096
248	RMSNorm_6_1	RMSNorm	1 × 46116 × 4096
249	Attention_6	Grouped Query Attn	1 × 46116 × 4096
250	Add_6_attn	Add	1 × 46116 × 4096
251	RMSNorm_6_2	RMSNorm	1 × 46116 × 4096
252	FFN_6	SwiGLU	1 × 46116 × 4096
253	Add_6_ffn	Add	1 × 46116 × 4096
254	RMSNorm_7_1	RMSNorm	1 × 46116 × 4096
255	Attention_7	Grouped Query Attn	1 × 46116 × 4096
256	Add_7_attn	Add	1 × 46116 × 4096
257	RMSNorm_7_2	RMSNorm	1 × 46116 × 4096
258	FFN_7	SwiGLU	1 × 46116 × 4096
259	Add_7_ffn	Add	1 × 46116 × 4096
260	RMSNorm_8_1	RMSNorm	1 × 46116 × 4096
261	Attention_8	Grouped Query Attn	1 × 46116 × 4096
262	Add_8_attn	Add	1 × 46116 × 4096
263	RMSNorm_8_2	RMSNorm	1 × 46116 × 4096
264	FFN_8	SwiGLU	1 × 46116 × 4096
265	Add_8_ffn	Add	1 × 46116 × 4096
266	RMSNorm_9_1	RMSNorm	1 × 46116 × 4096
267	Attention_9	Grouped Query Attn	1 × 46116 × 4096
268	Add_9_attn	Add	1 × 46116 × 4096
269	RMSNorm_9_2	RMSNorm	1 × 46116 × 4096
270	FFN_9	SwiGLU	1 × 46116 × 4096
271	Add_9_ffn	Add	1 × 46116 × 4096
272	RMSNorm_10_1	RMSNorm	1 × 46116 × 4096
273	Attention_10	Grouped Query Attn	1 × 46116 × 4096
274	Add_10_attn	Add	1 × 46116 × 4096
275	RMSNorm_10_2	RMSNorm	1 × 46116 × 4096
276	FFN_10	SwiGLU	1 × 46116 × 4096
277	Add_10_ffn	Add	1 × 46116 × 4096
278	RMSNorm_11_1	RMSNorm	1 × 46116 × 4096
279	Attention_11	Grouped Query Attn	1 × 46116 × 4096
280	Add_11_attn	Add	1 × 46116 × 4096
281	RMSNorm_11_2	RMSNorm	1 × 46116 × 4096
282	FFN_11	SwiGLU	1 × 46116 × 4096
283	Add_11_ffn	Add	1 × 46116 × 4096
284	RMSNorm_12_1	RMSNorm	1 × 46116 × 4096
285	Attention_12	Grouped Query Attn	1 × 46116 × 4096
286	Add_12_attn	Add	1 × 46116 × 4096
287	RMSNorm_12_2	RMSNorm	1 × 46116 × 4096
288	FFN_12	SwiGLU	1 × 46116 × 4096
289	Add_12_ffn	Add	1 × 46116 × 4096
290	RMSNorm_13_1	RMSNorm	1 × 46116 × 4096
291	Attention_13	Grouped Query Attn	1 × 46116 × 4096
292	Add_13_attn	Add	1 × 46116 × 4096
293	RMSNorm_13_2	RMSNorm	1 × 46116 × 4096
294	FFN_13	SwiGLU	1 × 46116 × 4096
295	Add_13_ffn	Add	1 × 46116 × 4096
296	RMSNorm_14_1	RMSNorm	1 × 46116 × 4096
297	Attention_14	Grouped Query Attn	1 × 46116 × 4096
298	Add_14_attn	Add	1 × 46116 × 4096
299	RMSNorm_14_2	RMSNorm	1 × 46116 × 4096
300	FFN_14	SwiGLU	1 × 46116 × 4096
301	Add_14_ffn	Add	1 × 46116 × 4096
302	RMSNorm_15_1	RMSNorm	1 × 46116 × 4096
303	Attention_15	Grouped Query Attn	1 × 46116 × 4096
304	Add_15_attn	Add	1 × 46116 × 4096
305	RMSNorm_15_2	RMSNorm	1 × 46116 × 4096
306	FFN_15	SwiGLU	1 × 46116 × 4096
307	Add_15_ffn	Add	1 × 46116 × 4096
308	RMSNorm_16_1	RMSNorm	1 × 46116 × 4096
309	Attention_16	Grouped Query Attn	1 × 46116 × 4096
310	Add_16_attn	Add	1 × 46116 × 4096
311	RMSNorm_16_2	RMSNorm	1 × 46116 × 4096
312	FFN_16	SwiGLU	1 × 46116 × 4096
313	Add_16_ffn	Add	1 × 46116 × 4096
314	RMSNorm_17_1	RMSNorm	1 × 46116 × 4096
315	Attention_17	Grouped Query Attn	1 × 46116 × 4096
316	Add_17_attn	Add	1 × 46116 × 4096
317	RMSNorm_17_2	RMSNorm	1 × 46116 × 4096
318	FFN_17	SwiGLU	1 × 46116 × 4096
319	Add_17_ffn	Add	1 × 46116 × 4096
320	RMSNorm_18_1	RMSNorm	1 × 46116 × 4096
321	Attention_18	Grouped Query Attn	1 × 46116 × 4096
322	Add_18_attn	Add	1 × 46116 × 4096
323	RMSNorm_18_2	RMSNorm	1 × 46116 × 4096
324	FFN_18	SwiGLU	1 × 46116 × 4096
325	Add_18_ffn	Add	1 × 46116 × 4096
326	RMSNorm_19_1	RMSNorm	1 × 46116 × 4096
327	Attention_19	Grouped Query Attn	1 × 46116 × 4096
328	Add_19_attn	Add	1 × 46116 × 4096
329	RMSNorm_19_2	RMSNorm	1 × 46116 × 4096
330	FFN_19	SwiGLU	1 × 46116 × 4096
331	Add_19_ffn	Add	1 × 46116 × 4096
332	RMSNorm_20_1	RMSNorm	1 × 46116 × 4096
333	Attention_20	Grouped Query Attn	1 × 46116 × 4096
334	Add_20_attn	Add	1 × 46116 × 4096
335	RMSNorm_20_2	RMSNorm	1 × 46116 × 4096
336	FFN_20	SwiGLU	1 × 46116 × 4096
337	Add_20_ffn	Add	1 × 46116 × 4096
338	RMSNorm_21_1	RMSNorm	1 × 46116 × 4096
339	Attention_21	Grouped Query Attn	1 × 46116 × 4096
340	Add_21_attn	Add	1 × 46116 × 4096
341	RMSNorm_21_2	RMSNorm	1 × 46116 × 4096
342	FFN_21	SwiGLU	1 × 46116 × 4096
343	Add_21_ffn	Add	1 × 46116 × 4096
344	RMSNorm_22_1	RMSNorm	1 × 46116 × 4096
345	Attention_22	Grouped Query Attn	1 × 46116 × 4096
346	Add_22_attn	Add	1 × 46116 × 4096
347	RMSNorm_22_2	RMSNorm	1 × 46116 × 4096
348	FFN_22	SwiGLU	1 × 46116 × 4096
349	Add_22_ffn	Add	1 × 46116 × 4096
350	RMSNorm_23_1	RMSNorm	1 × 46116 × 4096
351	Attention_23	Grouped Query Attn	1 × 46116 × 4096
352	Add_23_attn	Add	1 × 46116 × 4096
353	RMSNorm_23_2	RMSNorm	1 × 46116 × 4096
354	FFN_23	SwiGLU	1 × 46116 × 4096
355	Add_23_ffn	Add	1 × 46116 × 4096
356	RMSNorm_24_1	RMSNorm	1 × 46116 × 4096
357	Attention_24	Grouped Query Attn	1 × 46116 × 4096
358	Add_24_attn	Add	1 × 46116 × 4096
359	RMSNorm_24_2	RMSNorm	1 × 46116 × 4096
360	FFN_24	SwiGLU	1 × 46116 × 4096
361	Add_24_ffn	Add	1 × 46116 × 4096
362	RMSNorm_25_1	RMSNorm	1 × 46116 × 4096
363	Attention_25	Grouped Query Attn	1 × 46116 × 4096
364	Add_25_attn	Add	1 × 46116 × 4096
365	RMSNorm_25_2	RMSNorm	1 × 46116 × 4096
366	FFN_25	SwiGLU	1 × 46116 × 4096
367	Add_25_ffn	Add	1 × 46116 × 4096
368	RMSNorm_26_1	RMSNorm	1 × 46116 × 4096
369	Attention_26	Grouped Query Attn	1 × 46116 × 4096
370	Add_26_attn	Add	1 × 46116 × 4096
371	RMSNorm_26_2	RMSNorm	1 × 46116 × 4096
372	FFN_26	SwiGLU	1 × 46116 × 4096
373	Add_26_ffn	Add	1 × 46116 × 4096
374	RMSNorm_27_1	RMSNorm	1 × 46116 × 4096
375	Attention_27	Grouped Query Attn	1 × 46116 × 4096
376	Add_27_attn	Add	1 × 46116 × 4096
377	RMSNorm_27_2	RMSNorm	1 × 46116 × 4096
378	FFN_27	SwiGLU	1 × 46116 × 4096
379	Add_27_ffn	Add	1 × 46116 × 4096
380	RMSNorm_28_1	RMSNorm	1 × 46116 × 4096
381	Attention_28	Grouped Query Attn	1 × 46116 × 4096
382	Add_28_attn	Add	1 × 46116 × 4096
383	RMSNorm_28_2	RMSNorm	1 × 46116 × 4096
384	FFN_28	SwiGLU	1 × 46116 × 4096
385	Add_28_ffn	Add	1 × 46116 × 4096
386	RMSNorm_29_1	RMSNorm	1 × 46116 × 4096
387	Attention_29	Grouped Query Attn	1 × 46116 × 4096
388	Add_29_attn	Add	1 × 46116 × 4096
389	RMSNorm_29_2	RMSNorm	1 × 46116 × 4096
390	FFN_29	SwiGLU	1 × 46116 × 4096
391	Add_29_ffn	Add	1 × 46116 × 4096
392	RMSNorm_30_1	RMSNorm	1 × 46116 × 4096
393	Attention_30	Grouped Query Attn	1 × 46116 × 4096
394	Add_30_attn	Add	1 × 46116 × 4096
395	RMSNorm_30_2	RMSNorm	1 × 46116 × 4096
396	FFN_30	SwiGLU	1 × 46116 × 4096
397	Add_30_ffn	Add	1 × 46116 × 4096
398	RMSNorm_31_1	RMSNorm	1 × 46116 × 4096
399	Attention_31	Grouped Query Attn	1 × 46116 × 4096
400	Add_31_attn	Add	1 × 46116 × 4096
401	RMSNorm_31_2	RMSNorm	1 × 46116 × 4096
402	FFN_31	SwiGLU	1 × 46116 × 4096
403	Add_31_ffn	Add	1 × 46116 × 4096
404	RMSNorm_32_1	RMSNorm	1 × 46116 × 4096
405	Attention_32	Grouped Query Attn	1 × 46116 × 4096
406	Add_32_attn	Add	1 × 46116 × 4096
407	RMSNorm_32_2	RMSNorm	1 × 46116 × 4096
408	FFN_32	SwiGLU	1 × 46116 × 4096
409	Add_32_ffn	Add	1 × 46116 × 4096
410	RMSNorm_33_1	RMSNorm	1 × 46116 × 4096
411	Attention_33	Grouped Query Attn	1 × 46116 × 4096
412	Add_33_attn	Add	1 × 46116 × 4096
413	RMSNorm_33_2	RMSNorm	1 × 46116 × 4096
414	FFN_33	SwiGLU	1 × 46116 × 4096
415	Add_33_ffn	Add	1 × 46116 × 4096
416	RMSNorm_34_1	RMSNorm	1 × 46116 × 4096
417	Attention_34	Grouped Query Attn	1 × 46116 × 4096
418	Add_34_attn	Add	1 × 46116 × 4096
419	RMSNorm_34_2	RMSNorm	1 × 46116 × 4096
420	FFN_34	SwiGLU	1 × 46116 × 4096
421	Add_34_ffn	Add	1 × 46116 × 4096
422	RMSNorm_35_1	RMSNorm	1 × 46116 × 4096
423	Attention_35	Grouped Query Attn	1 × 46116 × 4096
424	Add_35_attn	Add	1 × 46116 × 4096
425	RMSNorm_35_2	RMSNorm	1 × 46116 × 4096
426	FFN_35	SwiGLU	1 × 46116 × 4096
427	Add_35_ffn	Add	1 × 46116 × 4096
428	RMSNorm_36_1	RMSNorm	1 × 46116 × 4096
429	Attention_36	Grouped Query Attn	1 × 46116 × 4096
430	Add_36_attn	Add	1 × 46116 × 4096
431	RMSNorm_36_2	RMSNorm	1 × 46116 × 4096
432	FFN_36	SwiGLU	1 × 46116 × 4096
433	Add_36_ffn	Add	1 × 46116 × 4096
434	Final_RMSNorm	RMSNorm	1 × 46116 × 4096
435	LM_Head	Linear	1 × 46116 × 151748
436	Output	Output	1 × 46116 × 151748

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

infoAt 87 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 openbmb/MiniCPM-o-4_5 --plan --share

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

Other minicpmo checkpoints

MiniCPM-o-2_6

8.34B derived · -3.85% against the checkpoint