N Neurarch Architectures Models Checks Data Docs Open the app

Models / nemotron_h

NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4

Reconstructed from its own config.json with no weights read. 726K 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
1496B
1,495,510,032,384 parameters
In the published checkpoint
67.23B
67,228,556,288 scalars · safetensors.total, read 2026-08-24
Delta
+2125%

quantized This checkpoint is stored quantized (modelopt). 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
532
Will it forward-pass
Yes
Priced on
A10G (24GB)
Est. one run
$1620896.04
CardMemory
T4 (16GB)weights + activationsdoes not fit
A100 (40GB)weights + activationsdoes not fit
H100 (80GB)weights + activationsdoes not fit

Structure

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

LayerTypeOutput shape
1InputInput1 × 262144
2EmbeddingEmbedding1 × 262144 × 4096
3RoPERoPE1 × 262144 × 4096
4LayerNorm_1_1LayerNorm1 × 262144 × 4096
5Attention_1Grouped Query Attn1 × 262144 × 4096
6Add_1_attnAdd1 × 262144 × 4096
7LayerNorm_1_2LayerNorm1 × 262144 × 4096
8MoE_1Shared-Expert MoE1 × 262144 × 4096
9Add_1_ffnAdd1 × 262144 × 4096
10LayerNorm_2_1LayerNorm1 × 262144 × 4096
11Attention_2Grouped Query Attn1 × 262144 × 4096
12Add_2_attnAdd1 × 262144 × 4096
13LayerNorm_2_2LayerNorm1 × 262144 × 4096
14MoE_2Shared-Expert MoE1 × 262144 × 4096
15Add_2_ffnAdd1 × 262144 × 4096
16LayerNorm_3_1LayerNorm1 × 262144 × 4096
17Attention_3Grouped Query Attn1 × 262144 × 4096
18Add_3_attnAdd1 × 262144 × 4096
19LayerNorm_3_2LayerNorm1 × 262144 × 4096
20MoE_3Shared-Expert MoE1 × 262144 × 4096
21Add_3_ffnAdd1 × 262144 × 4096
22LayerNorm_4_1LayerNorm1 × 262144 × 4096
23Attention_4Grouped Query Attn1 × 262144 × 4096
24Add_4_attnAdd1 × 262144 × 4096
25LayerNorm_4_2LayerNorm1 × 262144 × 4096
26MoE_4Shared-Expert MoE1 × 262144 × 4096
27Add_4_ffnAdd1 × 262144 × 4096
28LayerNorm_5_1LayerNorm1 × 262144 × 4096
29Attention_5Grouped Query Attn1 × 262144 × 4096
30Add_5_attnAdd1 × 262144 × 4096
31LayerNorm_5_2LayerNorm1 × 262144 × 4096
32MoE_5Shared-Expert MoE1 × 262144 × 4096
33Add_5_ffnAdd1 × 262144 × 4096
34LayerNorm_6_1LayerNorm1 × 262144 × 4096
35Attention_6Grouped Query Attn1 × 262144 × 4096
36Add_6_attnAdd1 × 262144 × 4096
37LayerNorm_6_2LayerNorm1 × 262144 × 4096
38MoE_6Shared-Expert MoE1 × 262144 × 4096
39Add_6_ffnAdd1 × 262144 × 4096
40LayerNorm_7_1LayerNorm1 × 262144 × 4096
41Attention_7Grouped Query Attn1 × 262144 × 4096
42Add_7_attnAdd1 × 262144 × 4096
43LayerNorm_7_2LayerNorm1 × 262144 × 4096
44MoE_7Shared-Expert MoE1 × 262144 × 4096
45Add_7_ffnAdd1 × 262144 × 4096
46LayerNorm_8_1LayerNorm1 × 262144 × 4096
47Attention_8Grouped Query Attn1 × 262144 × 4096
48Add_8_attnAdd1 × 262144 × 4096
49LayerNorm_8_2LayerNorm1 × 262144 × 4096
50MoE_8Shared-Expert MoE1 × 262144 × 4096
51Add_8_ffnAdd1 × 262144 × 4096
52LayerNorm_9_1LayerNorm1 × 262144 × 4096
53Attention_9Grouped Query Attn1 × 262144 × 4096
54Add_9_attnAdd1 × 262144 × 4096
55LayerNorm_9_2LayerNorm1 × 262144 × 4096
56MoE_9Shared-Expert MoE1 × 262144 × 4096
57Add_9_ffnAdd1 × 262144 × 4096
58LayerNorm_10_1LayerNorm1 × 262144 × 4096
59Attention_10Grouped Query Attn1 × 262144 × 4096
60Add_10_attnAdd1 × 262144 × 4096
61LayerNorm_10_2LayerNorm1 × 262144 × 4096
62MoE_10Shared-Expert MoE1 × 262144 × 4096
63Add_10_ffnAdd1 × 262144 × 4096
64LayerNorm_11_1LayerNorm1 × 262144 × 4096
65Attention_11Grouped Query Attn1 × 262144 × 4096
66Add_11_attnAdd1 × 262144 × 4096
67LayerNorm_11_2LayerNorm1 × 262144 × 4096
68MoE_11Shared-Expert MoE1 × 262144 × 4096
69Add_11_ffnAdd1 × 262144 × 4096
70LayerNorm_12_1LayerNorm1 × 262144 × 4096
71Attention_12Grouped Query Attn1 × 262144 × 4096
72Add_12_attnAdd1 × 262144 × 4096
73LayerNorm_12_2LayerNorm1 × 262144 × 4096
74MoE_12Shared-Expert MoE1 × 262144 × 4096
75Add_12_ffnAdd1 × 262144 × 4096
76LayerNorm_13_1LayerNorm1 × 262144 × 4096
77Attention_13Grouped Query Attn1 × 262144 × 4096
78Add_13_attnAdd1 × 262144 × 4096
79LayerNorm_13_2LayerNorm1 × 262144 × 4096
80MoE_13Shared-Expert MoE1 × 262144 × 4096
81Add_13_ffnAdd1 × 262144 × 4096
82LayerNorm_14_1LayerNorm1 × 262144 × 4096
83Attention_14Grouped Query Attn1 × 262144 × 4096
84Add_14_attnAdd1 × 262144 × 4096
85LayerNorm_14_2LayerNorm1 × 262144 × 4096
86MoE_14Shared-Expert MoE1 × 262144 × 4096
87Add_14_ffnAdd1 × 262144 × 4096
88LayerNorm_15_1LayerNorm1 × 262144 × 4096
89Attention_15Grouped Query Attn1 × 262144 × 4096
90Add_15_attnAdd1 × 262144 × 4096
91LayerNorm_15_2LayerNorm1 × 262144 × 4096
92MoE_15Shared-Expert MoE1 × 262144 × 4096
93Add_15_ffnAdd1 × 262144 × 4096
94LayerNorm_16_1LayerNorm1 × 262144 × 4096
95Attention_16Grouped Query Attn1 × 262144 × 4096
96Add_16_attnAdd1 × 262144 × 4096
97LayerNorm_16_2LayerNorm1 × 262144 × 4096
98MoE_16Shared-Expert MoE1 × 262144 × 4096
99Add_16_ffnAdd1 × 262144 × 4096
100LayerNorm_17_1LayerNorm1 × 262144 × 4096
101Attention_17Grouped Query Attn1 × 262144 × 4096
102Add_17_attnAdd1 × 262144 × 4096
103LayerNorm_17_2LayerNorm1 × 262144 × 4096
104MoE_17Shared-Expert MoE1 × 262144 × 4096
105Add_17_ffnAdd1 × 262144 × 4096
106LayerNorm_18_1LayerNorm1 × 262144 × 4096
107Attention_18Grouped Query Attn1 × 262144 × 4096
108Add_18_attnAdd1 × 262144 × 4096
109LayerNorm_18_2LayerNorm1 × 262144 × 4096
110MoE_18Shared-Expert MoE1 × 262144 × 4096
111Add_18_ffnAdd1 × 262144 × 4096
112LayerNorm_19_1LayerNorm1 × 262144 × 4096
113Attention_19Grouped Query Attn1 × 262144 × 4096
114Add_19_attnAdd1 × 262144 × 4096
115LayerNorm_19_2LayerNorm1 × 262144 × 4096
116MoE_19Shared-Expert MoE1 × 262144 × 4096
117Add_19_ffnAdd1 × 262144 × 4096
118LayerNorm_20_1LayerNorm1 × 262144 × 4096
119Attention_20Grouped Query Attn1 × 262144 × 4096
120Add_20_attnAdd1 × 262144 × 4096
121LayerNorm_20_2LayerNorm1 × 262144 × 4096
122MoE_20Shared-Expert MoE1 × 262144 × 4096
123Add_20_ffnAdd1 × 262144 × 4096
124LayerNorm_21_1LayerNorm1 × 262144 × 4096
125Attention_21Grouped Query Attn1 × 262144 × 4096
126Add_21_attnAdd1 × 262144 × 4096
127LayerNorm_21_2LayerNorm1 × 262144 × 4096
128MoE_21Shared-Expert MoE1 × 262144 × 4096
129Add_21_ffnAdd1 × 262144 × 4096
130LayerNorm_22_1LayerNorm1 × 262144 × 4096
131Attention_22Grouped Query Attn1 × 262144 × 4096
132Add_22_attnAdd1 × 262144 × 4096
133LayerNorm_22_2LayerNorm1 × 262144 × 4096
134MoE_22Shared-Expert MoE1 × 262144 × 4096
135Add_22_ffnAdd1 × 262144 × 4096
136LayerNorm_23_1LayerNorm1 × 262144 × 4096
137Attention_23Grouped Query Attn1 × 262144 × 4096
138Add_23_attnAdd1 × 262144 × 4096
139LayerNorm_23_2LayerNorm1 × 262144 × 4096
140MoE_23Shared-Expert MoE1 × 262144 × 4096
141Add_23_ffnAdd1 × 262144 × 4096
142LayerNorm_24_1LayerNorm1 × 262144 × 4096
143Attention_24Grouped Query Attn1 × 262144 × 4096
144Add_24_attnAdd1 × 262144 × 4096
145LayerNorm_24_2LayerNorm1 × 262144 × 4096
146MoE_24Shared-Expert MoE1 × 262144 × 4096
147Add_24_ffnAdd1 × 262144 × 4096
148LayerNorm_25_1LayerNorm1 × 262144 × 4096
149Attention_25Grouped Query Attn1 × 262144 × 4096
150Add_25_attnAdd1 × 262144 × 4096
151LayerNorm_25_2LayerNorm1 × 262144 × 4096
152MoE_25Shared-Expert MoE1 × 262144 × 4096
153Add_25_ffnAdd1 × 262144 × 4096
154LayerNorm_26_1LayerNorm1 × 262144 × 4096
155Attention_26Grouped Query Attn1 × 262144 × 4096
156Add_26_attnAdd1 × 262144 × 4096
157LayerNorm_26_2LayerNorm1 × 262144 × 4096
158MoE_26Shared-Expert MoE1 × 262144 × 4096
159Add_26_ffnAdd1 × 262144 × 4096
160LayerNorm_27_1LayerNorm1 × 262144 × 4096
161Attention_27Grouped Query Attn1 × 262144 × 4096
162Add_27_attnAdd1 × 262144 × 4096
163LayerNorm_27_2LayerNorm1 × 262144 × 4096
164MoE_27Shared-Expert MoE1 × 262144 × 4096
165Add_27_ffnAdd1 × 262144 × 4096
166LayerNorm_28_1LayerNorm1 × 262144 × 4096
167Attention_28Grouped Query Attn1 × 262144 × 4096
168Add_28_attnAdd1 × 262144 × 4096
169LayerNorm_28_2LayerNorm1 × 262144 × 4096
170MoE_28Shared-Expert MoE1 × 262144 × 4096
171Add_28_ffnAdd1 × 262144 × 4096
172LayerNorm_29_1LayerNorm1 × 262144 × 4096
173Attention_29Grouped Query Attn1 × 262144 × 4096
174Add_29_attnAdd1 × 262144 × 4096
175LayerNorm_29_2LayerNorm1 × 262144 × 4096
176MoE_29Shared-Expert MoE1 × 262144 × 4096
177Add_29_ffnAdd1 × 262144 × 4096
178LayerNorm_30_1LayerNorm1 × 262144 × 4096
179Attention_30Grouped Query Attn1 × 262144 × 4096
180Add_30_attnAdd1 × 262144 × 4096
181LayerNorm_30_2LayerNorm1 × 262144 × 4096
182MoE_30Shared-Expert MoE1 × 262144 × 4096
183Add_30_ffnAdd1 × 262144 × 4096
184LayerNorm_31_1LayerNorm1 × 262144 × 4096
185Attention_31Grouped Query Attn1 × 262144 × 4096
186Add_31_attnAdd1 × 262144 × 4096
187LayerNorm_31_2LayerNorm1 × 262144 × 4096
188MoE_31Shared-Expert MoE1 × 262144 × 4096
189Add_31_ffnAdd1 × 262144 × 4096
190LayerNorm_32_1LayerNorm1 × 262144 × 4096
191Attention_32Grouped Query Attn1 × 262144 × 4096
192Add_32_attnAdd1 × 262144 × 4096
193LayerNorm_32_2LayerNorm1 × 262144 × 4096
194MoE_32Shared-Expert MoE1 × 262144 × 4096
195Add_32_ffnAdd1 × 262144 × 4096
196LayerNorm_33_1LayerNorm1 × 262144 × 4096
197Attention_33Grouped Query Attn1 × 262144 × 4096
198Add_33_attnAdd1 × 262144 × 4096
199LayerNorm_33_2LayerNorm1 × 262144 × 4096
200MoE_33Shared-Expert MoE1 × 262144 × 4096
201Add_33_ffnAdd1 × 262144 × 4096
202LayerNorm_34_1LayerNorm1 × 262144 × 4096
203Attention_34Grouped Query Attn1 × 262144 × 4096
204Add_34_attnAdd1 × 262144 × 4096
205LayerNorm_34_2LayerNorm1 × 262144 × 4096
206MoE_34Shared-Expert MoE1 × 262144 × 4096
207Add_34_ffnAdd1 × 262144 × 4096
208LayerNorm_35_1LayerNorm1 × 262144 × 4096
209Attention_35Grouped Query Attn1 × 262144 × 4096
210Add_35_attnAdd1 × 262144 × 4096
211LayerNorm_35_2LayerNorm1 × 262144 × 4096
212MoE_35Shared-Expert MoE1 × 262144 × 4096
213Add_35_ffnAdd1 × 262144 × 4096
214LayerNorm_36_1LayerNorm1 × 262144 × 4096
215Attention_36Grouped Query Attn1 × 262144 × 4096
216Add_36_attnAdd1 × 262144 × 4096
217LayerNorm_36_2LayerNorm1 × 262144 × 4096
218MoE_36Shared-Expert MoE1 × 262144 × 4096
219Add_36_ffnAdd1 × 262144 × 4096
220LayerNorm_37_1LayerNorm1 × 262144 × 4096
221Attention_37Grouped Query Attn1 × 262144 × 4096
222Add_37_attnAdd1 × 262144 × 4096
223LayerNorm_37_2LayerNorm1 × 262144 × 4096
224MoE_37Shared-Expert MoE1 × 262144 × 4096
225Add_37_ffnAdd1 × 262144 × 4096
226LayerNorm_38_1LayerNorm1 × 262144 × 4096
227Attention_38Grouped Query Attn1 × 262144 × 4096
228Add_38_attnAdd1 × 262144 × 4096
229LayerNorm_38_2LayerNorm1 × 262144 × 4096
230MoE_38Shared-Expert MoE1 × 262144 × 4096
231Add_38_ffnAdd1 × 262144 × 4096
232LayerNorm_39_1LayerNorm1 × 262144 × 4096
233Attention_39Grouped Query Attn1 × 262144 × 4096
234Add_39_attnAdd1 × 262144 × 4096
235LayerNorm_39_2LayerNorm1 × 262144 × 4096
236MoE_39Shared-Expert MoE1 × 262144 × 4096
237Add_39_ffnAdd1 × 262144 × 4096
238LayerNorm_40_1LayerNorm1 × 262144 × 4096
239Attention_40Grouped Query Attn1 × 262144 × 4096
240Add_40_attnAdd1 × 262144 × 4096
241LayerNorm_40_2LayerNorm1 × 262144 × 4096
242MoE_40Shared-Expert MoE1 × 262144 × 4096
243Add_40_ffnAdd1 × 262144 × 4096
244LayerNorm_41_1LayerNorm1 × 262144 × 4096
245Attention_41Grouped Query Attn1 × 262144 × 4096
246Add_41_attnAdd1 × 262144 × 4096
247LayerNorm_41_2LayerNorm1 × 262144 × 4096
248MoE_41Shared-Expert MoE1 × 262144 × 4096
249Add_41_ffnAdd1 × 262144 × 4096
250LayerNorm_42_1LayerNorm1 × 262144 × 4096
251Attention_42Grouped Query Attn1 × 262144 × 4096
252Add_42_attnAdd1 × 262144 × 4096
253LayerNorm_42_2LayerNorm1 × 262144 × 4096
254MoE_42Shared-Expert MoE1 × 262144 × 4096
255Add_42_ffnAdd1 × 262144 × 4096
256LayerNorm_43_1LayerNorm1 × 262144 × 4096
257Attention_43Grouped Query Attn1 × 262144 × 4096
258Add_43_attnAdd1 × 262144 × 4096
259LayerNorm_43_2LayerNorm1 × 262144 × 4096
260MoE_43Shared-Expert MoE1 × 262144 × 4096
261Add_43_ffnAdd1 × 262144 × 4096
262LayerNorm_44_1LayerNorm1 × 262144 × 4096
263Attention_44Grouped Query Attn1 × 262144 × 4096
264Add_44_attnAdd1 × 262144 × 4096
265LayerNorm_44_2LayerNorm1 × 262144 × 4096
266MoE_44Shared-Expert MoE1 × 262144 × 4096
267Add_44_ffnAdd1 × 262144 × 4096
268LayerNorm_45_1LayerNorm1 × 262144 × 4096
269Attention_45Grouped Query Attn1 × 262144 × 4096
270Add_45_attnAdd1 × 262144 × 4096
271LayerNorm_45_2LayerNorm1 × 262144 × 4096
272MoE_45Shared-Expert MoE1 × 262144 × 4096
273Add_45_ffnAdd1 × 262144 × 4096
274LayerNorm_46_1LayerNorm1 × 262144 × 4096
275Attention_46Grouped Query Attn1 × 262144 × 4096
276Add_46_attnAdd1 × 262144 × 4096
277LayerNorm_46_2LayerNorm1 × 262144 × 4096
278MoE_46Shared-Expert MoE1 × 262144 × 4096
279Add_46_ffnAdd1 × 262144 × 4096
280LayerNorm_47_1LayerNorm1 × 262144 × 4096
281Attention_47Grouped Query Attn1 × 262144 × 4096
282Add_47_attnAdd1 × 262144 × 4096
283LayerNorm_47_2LayerNorm1 × 262144 × 4096
284MoE_47Shared-Expert MoE1 × 262144 × 4096
285Add_47_ffnAdd1 × 262144 × 4096
286LayerNorm_48_1LayerNorm1 × 262144 × 4096
287Attention_48Grouped Query Attn1 × 262144 × 4096
288Add_48_attnAdd1 × 262144 × 4096
289LayerNorm_48_2LayerNorm1 × 262144 × 4096
290MoE_48Shared-Expert MoE1 × 262144 × 4096
291Add_48_ffnAdd1 × 262144 × 4096
292LayerNorm_49_1LayerNorm1 × 262144 × 4096
293Attention_49Grouped Query Attn1 × 262144 × 4096
294Add_49_attnAdd1 × 262144 × 4096
295LayerNorm_49_2LayerNorm1 × 262144 × 4096
296MoE_49Shared-Expert MoE1 × 262144 × 4096
297Add_49_ffnAdd1 × 262144 × 4096
298LayerNorm_50_1LayerNorm1 × 262144 × 4096
299Attention_50Grouped Query Attn1 × 262144 × 4096
300Add_50_attnAdd1 × 262144 × 4096
301LayerNorm_50_2LayerNorm1 × 262144 × 4096
302MoE_50Shared-Expert MoE1 × 262144 × 4096
303Add_50_ffnAdd1 × 262144 × 4096
304LayerNorm_51_1LayerNorm1 × 262144 × 4096
305Attention_51Grouped Query Attn1 × 262144 × 4096
306Add_51_attnAdd1 × 262144 × 4096
307LayerNorm_51_2LayerNorm1 × 262144 × 4096
308MoE_51Shared-Expert MoE1 × 262144 × 4096
309Add_51_ffnAdd1 × 262144 × 4096
310LayerNorm_52_1LayerNorm1 × 262144 × 4096
311Attention_52Grouped Query Attn1 × 262144 × 4096
312Add_52_attnAdd1 × 262144 × 4096
313LayerNorm_52_2LayerNorm1 × 262144 × 4096
314MoE_52Shared-Expert MoE1 × 262144 × 4096
315Add_52_ffnAdd1 × 262144 × 4096
316LayerNorm_53_1LayerNorm1 × 262144 × 4096
317Attention_53Grouped Query Attn1 × 262144 × 4096
318Add_53_attnAdd1 × 262144 × 4096
319LayerNorm_53_2LayerNorm1 × 262144 × 4096
320MoE_53Shared-Expert MoE1 × 262144 × 4096
321Add_53_ffnAdd1 × 262144 × 4096
322LayerNorm_54_1LayerNorm1 × 262144 × 4096
323Attention_54Grouped Query Attn1 × 262144 × 4096
324Add_54_attnAdd1 × 262144 × 4096
325LayerNorm_54_2LayerNorm1 × 262144 × 4096
326MoE_54Shared-Expert MoE1 × 262144 × 4096
327Add_54_ffnAdd1 × 262144 × 4096
328LayerNorm_55_1LayerNorm1 × 262144 × 4096
329Attention_55Grouped Query Attn1 × 262144 × 4096
330Add_55_attnAdd1 × 262144 × 4096
331LayerNorm_55_2LayerNorm1 × 262144 × 4096
332MoE_55Shared-Expert MoE1 × 262144 × 4096
333Add_55_ffnAdd1 × 262144 × 4096
334LayerNorm_56_1LayerNorm1 × 262144 × 4096
335Attention_56Grouped Query Attn1 × 262144 × 4096
336Add_56_attnAdd1 × 262144 × 4096
337LayerNorm_56_2LayerNorm1 × 262144 × 4096
338MoE_56Shared-Expert MoE1 × 262144 × 4096
339Add_56_ffnAdd1 × 262144 × 4096
340LayerNorm_57_1LayerNorm1 × 262144 × 4096
341Attention_57Grouped Query Attn1 × 262144 × 4096
342Add_57_attnAdd1 × 262144 × 4096
343LayerNorm_57_2LayerNorm1 × 262144 × 4096
344MoE_57Shared-Expert MoE1 × 262144 × 4096
345Add_57_ffnAdd1 × 262144 × 4096
346LayerNorm_58_1LayerNorm1 × 262144 × 4096
347Attention_58Grouped Query Attn1 × 262144 × 4096
348Add_58_attnAdd1 × 262144 × 4096
349LayerNorm_58_2LayerNorm1 × 262144 × 4096
350MoE_58Shared-Expert MoE1 × 262144 × 4096
351Add_58_ffnAdd1 × 262144 × 4096
352LayerNorm_59_1LayerNorm1 × 262144 × 4096
353Attention_59Grouped Query Attn1 × 262144 × 4096
354Add_59_attnAdd1 × 262144 × 4096
355LayerNorm_59_2LayerNorm1 × 262144 × 4096
356MoE_59Shared-Expert MoE1 × 262144 × 4096
357Add_59_ffnAdd1 × 262144 × 4096
358LayerNorm_60_1LayerNorm1 × 262144 × 4096
359Attention_60Grouped Query Attn1 × 262144 × 4096
360Add_60_attnAdd1 × 262144 × 4096
361LayerNorm_60_2LayerNorm1 × 262144 × 4096
362MoE_60Shared-Expert MoE1 × 262144 × 4096
363Add_60_ffnAdd1 × 262144 × 4096
364LayerNorm_61_1LayerNorm1 × 262144 × 4096
365Attention_61Grouped Query Attn1 × 262144 × 4096
366Add_61_attnAdd1 × 262144 × 4096
367LayerNorm_61_2LayerNorm1 × 262144 × 4096
368MoE_61Shared-Expert MoE1 × 262144 × 4096
369Add_61_ffnAdd1 × 262144 × 4096
370LayerNorm_62_1LayerNorm1 × 262144 × 4096
371Attention_62Grouped Query Attn1 × 262144 × 4096
372Add_62_attnAdd1 × 262144 × 4096
373LayerNorm_62_2LayerNorm1 × 262144 × 4096
374MoE_62Shared-Expert MoE1 × 262144 × 4096
375Add_62_ffnAdd1 × 262144 × 4096
376LayerNorm_63_1LayerNorm1 × 262144 × 4096
377Attention_63Grouped Query Attn1 × 262144 × 4096
378Add_63_attnAdd1 × 262144 × 4096
379LayerNorm_63_2LayerNorm1 × 262144 × 4096
380MoE_63Shared-Expert MoE1 × 262144 × 4096
381Add_63_ffnAdd1 × 262144 × 4096
382LayerNorm_64_1LayerNorm1 × 262144 × 4096
383Attention_64Grouped Query Attn1 × 262144 × 4096
384Add_64_attnAdd1 × 262144 × 4096
385LayerNorm_64_2LayerNorm1 × 262144 × 4096
386MoE_64Shared-Expert MoE1 × 262144 × 4096
387Add_64_ffnAdd1 × 262144 × 4096
388LayerNorm_65_1LayerNorm1 × 262144 × 4096
389Attention_65Grouped Query Attn1 × 262144 × 4096
390Add_65_attnAdd1 × 262144 × 4096
391LayerNorm_65_2LayerNorm1 × 262144 × 4096
392MoE_65Shared-Expert MoE1 × 262144 × 4096
393Add_65_ffnAdd1 × 262144 × 4096
394LayerNorm_66_1LayerNorm1 × 262144 × 4096
395Attention_66Grouped Query Attn1 × 262144 × 4096
396Add_66_attnAdd1 × 262144 × 4096
397LayerNorm_66_2LayerNorm1 × 262144 × 4096
398MoE_66Shared-Expert MoE1 × 262144 × 4096
399Add_66_ffnAdd1 × 262144 × 4096
400LayerNorm_67_1LayerNorm1 × 262144 × 4096
401Attention_67Grouped Query Attn1 × 262144 × 4096
402Add_67_attnAdd1 × 262144 × 4096
403LayerNorm_67_2LayerNorm1 × 262144 × 4096
404MoE_67Shared-Expert MoE1 × 262144 × 4096
405Add_67_ffnAdd1 × 262144 × 4096
406LayerNorm_68_1LayerNorm1 × 262144 × 4096
407Attention_68Grouped Query Attn1 × 262144 × 4096
408Add_68_attnAdd1 × 262144 × 4096
409LayerNorm_68_2LayerNorm1 × 262144 × 4096
410MoE_68Shared-Expert MoE1 × 262144 × 4096
411Add_68_ffnAdd1 × 262144 × 4096
412LayerNorm_69_1LayerNorm1 × 262144 × 4096
413Attention_69Grouped Query Attn1 × 262144 × 4096
414Add_69_attnAdd1 × 262144 × 4096
415LayerNorm_69_2LayerNorm1 × 262144 × 4096
416MoE_69Shared-Expert MoE1 × 262144 × 4096
417Add_69_ffnAdd1 × 262144 × 4096
418LayerNorm_70_1LayerNorm1 × 262144 × 4096
419Attention_70Grouped Query Attn1 × 262144 × 4096
420Add_70_attnAdd1 × 262144 × 4096
421LayerNorm_70_2LayerNorm1 × 262144 × 4096
422MoE_70Shared-Expert MoE1 × 262144 × 4096
423Add_70_ffnAdd1 × 262144 × 4096
424LayerNorm_71_1LayerNorm1 × 262144 × 4096
425Attention_71Grouped Query Attn1 × 262144 × 4096
426Add_71_attnAdd1 × 262144 × 4096
427LayerNorm_71_2LayerNorm1 × 262144 × 4096
428MoE_71Shared-Expert MoE1 × 262144 × 4096
429Add_71_ffnAdd1 × 262144 × 4096
430LayerNorm_72_1LayerNorm1 × 262144 × 4096
431Attention_72Grouped Query Attn1 × 262144 × 4096
432Add_72_attnAdd1 × 262144 × 4096
433LayerNorm_72_2LayerNorm1 × 262144 × 4096
434MoE_72Shared-Expert MoE1 × 262144 × 4096
435Add_72_ffnAdd1 × 262144 × 4096
436LayerNorm_73_1LayerNorm1 × 262144 × 4096
437Attention_73Grouped Query Attn1 × 262144 × 4096
438Add_73_attnAdd1 × 262144 × 4096
439LayerNorm_73_2LayerNorm1 × 262144 × 4096
440MoE_73Shared-Expert MoE1 × 262144 × 4096
441Add_73_ffnAdd1 × 262144 × 4096
442LayerNorm_74_1LayerNorm1 × 262144 × 4096
443Attention_74Grouped Query Attn1 × 262144 × 4096
444Add_74_attnAdd1 × 262144 × 4096
445LayerNorm_74_2LayerNorm1 × 262144 × 4096
446MoE_74Shared-Expert MoE1 × 262144 × 4096
447Add_74_ffnAdd1 × 262144 × 4096
448LayerNorm_75_1LayerNorm1 × 262144 × 4096
449Attention_75Grouped Query Attn1 × 262144 × 4096
450Add_75_attnAdd1 × 262144 × 4096
451LayerNorm_75_2LayerNorm1 × 262144 × 4096
452MoE_75Shared-Expert MoE1 × 262144 × 4096
453Add_75_ffnAdd1 × 262144 × 4096
454LayerNorm_76_1LayerNorm1 × 262144 × 4096
455Attention_76Grouped Query Attn1 × 262144 × 4096
456Add_76_attnAdd1 × 262144 × 4096
457LayerNorm_76_2LayerNorm1 × 262144 × 4096
458MoE_76Shared-Expert MoE1 × 262144 × 4096
459Add_76_ffnAdd1 × 262144 × 4096
460LayerNorm_77_1LayerNorm1 × 262144 × 4096
461Attention_77Grouped Query Attn1 × 262144 × 4096
462Add_77_attnAdd1 × 262144 × 4096
463LayerNorm_77_2LayerNorm1 × 262144 × 4096
464MoE_77Shared-Expert MoE1 × 262144 × 4096
465Add_77_ffnAdd1 × 262144 × 4096
466LayerNorm_78_1LayerNorm1 × 262144 × 4096
467Attention_78Grouped Query Attn1 × 262144 × 4096
468Add_78_attnAdd1 × 262144 × 4096
469LayerNorm_78_2LayerNorm1 × 262144 × 4096
470MoE_78Shared-Expert MoE1 × 262144 × 4096
471Add_78_ffnAdd1 × 262144 × 4096
472LayerNorm_79_1LayerNorm1 × 262144 × 4096
473Attention_79Grouped Query Attn1 × 262144 × 4096
474Add_79_attnAdd1 × 262144 × 4096
475LayerNorm_79_2LayerNorm1 × 262144 × 4096
476MoE_79Shared-Expert MoE1 × 262144 × 4096
477Add_79_ffnAdd1 × 262144 × 4096
478LayerNorm_80_1LayerNorm1 × 262144 × 4096
479Attention_80Grouped Query Attn1 × 262144 × 4096
480Add_80_attnAdd1 × 262144 × 4096
481LayerNorm_80_2LayerNorm1 × 262144 × 4096
482MoE_80Shared-Expert MoE1 × 262144 × 4096
483Add_80_ffnAdd1 × 262144 × 4096
484LayerNorm_81_1LayerNorm1 × 262144 × 4096
485Attention_81Grouped Query Attn1 × 262144 × 4096
486Add_81_attnAdd1 × 262144 × 4096
487LayerNorm_81_2LayerNorm1 × 262144 × 4096
488MoE_81Shared-Expert MoE1 × 262144 × 4096
489Add_81_ffnAdd1 × 262144 × 4096
490LayerNorm_82_1LayerNorm1 × 262144 × 4096
491Attention_82Grouped Query Attn1 × 262144 × 4096
492Add_82_attnAdd1 × 262144 × 4096
493LayerNorm_82_2LayerNorm1 × 262144 × 4096
494MoE_82Shared-Expert MoE1 × 262144 × 4096
495Add_82_ffnAdd1 × 262144 × 4096
496LayerNorm_83_1LayerNorm1 × 262144 × 4096
497Attention_83Grouped Query Attn1 × 262144 × 4096
498Add_83_attnAdd1 × 262144 × 4096
499LayerNorm_83_2LayerNorm1 × 262144 × 4096
500MoE_83Shared-Expert MoE1 × 262144 × 4096
501Add_83_ffnAdd1 × 262144 × 4096
502LayerNorm_84_1LayerNorm1 × 262144 × 4096
503Attention_84Grouped Query Attn1 × 262144 × 4096
504Add_84_attnAdd1 × 262144 × 4096
505LayerNorm_84_2LayerNorm1 × 262144 × 4096
506MoE_84Shared-Expert MoE1 × 262144 × 4096
507Add_84_ffnAdd1 × 262144 × 4096
508LayerNorm_85_1LayerNorm1 × 262144 × 4096
509Attention_85Grouped Query Attn1 × 262144 × 4096
510Add_85_attnAdd1 × 262144 × 4096
511LayerNorm_85_2LayerNorm1 × 262144 × 4096
512MoE_85Shared-Expert MoE1 × 262144 × 4096
513Add_85_ffnAdd1 × 262144 × 4096
514LayerNorm_86_1LayerNorm1 × 262144 × 4096
515Attention_86Grouped Query Attn1 × 262144 × 4096
516Add_86_attnAdd1 × 262144 × 4096
517LayerNorm_86_2LayerNorm1 × 262144 × 4096
518MoE_86Shared-Expert MoE1 × 262144 × 4096
519Add_86_ffnAdd1 × 262144 × 4096
520LayerNorm_87_1LayerNorm1 × 262144 × 4096
521Attention_87Grouped Query Attn1 × 262144 × 4096
522Add_87_attnAdd1 × 262144 × 4096
523LayerNorm_87_2LayerNorm1 × 262144 × 4096
524MoE_87Shared-Expert MoE1 × 262144 × 4096
525Add_87_ffnAdd1 × 262144 × 4096
526LayerNorm_88_1LayerNorm1 × 262144 × 4096
527Attention_88Grouped Query Attn1 × 262144 × 4096
528Add_88_attnAdd1 × 262144 × 4096
529LayerNorm_88_2LayerNorm1 × 262144 × 4096
530MoE_88Shared-Expert MoE1 × 262144 × 4096
531Add_88_ffnAdd1 × 262144 × 4096
532Final_LayerNormLayerNorm1 × 262144 × 4096
533LM_HeadLinear1 × 262144 × 131072
534OutputOutput1 × 262144 × 131072

What the verifier says

infoAt 88 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 nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 --plan --share

Other nemotron_h checkpoints

NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
102B derived · +224% against the checkpoint
NVIDIA-Nemotron-3-Nano-30B-A3B-FP8
102B derived · +224% against the checkpoint
NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4
102B derived · +461% against the checkpoint
NVIDIA-Nemotron-3-Super-120B-A12B-BF16
1496B derived · +1110% against the checkpoint