Models / NemotronH_Nano_Omni_Reasoning_V3

Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4

Reconstructed from its own config.json with no weights read. 746K 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

3.08B

3,079,759,616 parameters

In the published checkpoint

18.33B

18,326,275,008 scalars · safetensors.total, read 2026-08-24

Delta

-83.2%

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

210

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$379336.61

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

Structure

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

	Layer	Type	Output shape
1	Input	Input	1 × 262144
2	Embedding	Embedding	1 × 262144 × 2688
3	Positional_Embedding	Learned Pos Embed	1 × 262144 × 2688
4	Attention_1	Multi-Head Attention	1 × 262144 × 2688
5	Add_1	Add	1 × 262144 × 2688
6	LayerNorm_1_1	LayerNorm	1 × 262144 × 2688
7	FFN_1	Feed Forward	1 × 262144 × 2688
8	Attention_2	Multi-Head Attention	1 × 262144 × 2688
9	Add_2	Add	1 × 262144 × 2688
10	LayerNorm_2_1	LayerNorm	1 × 262144 × 2688
11	FFN_2	Feed Forward	1 × 262144 × 2688
12	Attention_3	Multi-Head Attention	1 × 262144 × 2688
13	Add_3	Add	1 × 262144 × 2688
14	LayerNorm_3_1	LayerNorm	1 × 262144 × 2688
15	FFN_3	Feed Forward	1 × 262144 × 2688
16	Attention_4	Multi-Head Attention	1 × 262144 × 2688
17	Add_4	Add	1 × 262144 × 2688
18	LayerNorm_4_1	LayerNorm	1 × 262144 × 2688
19	FFN_4	Feed Forward	1 × 262144 × 2688
20	Attention_5	Multi-Head Attention	1 × 262144 × 2688
21	Add_5	Add	1 × 262144 × 2688
22	LayerNorm_5_1	LayerNorm	1 × 262144 × 2688
23	FFN_5	Feed Forward	1 × 262144 × 2688
24	Attention_6	Multi-Head Attention	1 × 262144 × 2688
25	Add_6	Add	1 × 262144 × 2688
26	LayerNorm_6_1	LayerNorm	1 × 262144 × 2688
27	FFN_6	Feed Forward	1 × 262144 × 2688
28	Attention_7	Multi-Head Attention	1 × 262144 × 2688
29	Add_7	Add	1 × 262144 × 2688
30	LayerNorm_7_1	LayerNorm	1 × 262144 × 2688
31	FFN_7	Feed Forward	1 × 262144 × 2688
32	Attention_8	Multi-Head Attention	1 × 262144 × 2688
33	Add_8	Add	1 × 262144 × 2688
34	LayerNorm_8_1	LayerNorm	1 × 262144 × 2688
35	FFN_8	Feed Forward	1 × 262144 × 2688
36	Attention_9	Multi-Head Attention	1 × 262144 × 2688
37	Add_9	Add	1 × 262144 × 2688
38	LayerNorm_9_1	LayerNorm	1 × 262144 × 2688
39	FFN_9	Feed Forward	1 × 262144 × 2688
40	Attention_10	Multi-Head Attention	1 × 262144 × 2688
41	Add_10	Add	1 × 262144 × 2688
42	LayerNorm_10_1	LayerNorm	1 × 262144 × 2688
43	FFN_10	Feed Forward	1 × 262144 × 2688
44	Attention_11	Multi-Head Attention	1 × 262144 × 2688
45	Add_11	Add	1 × 262144 × 2688
46	LayerNorm_11_1	LayerNorm	1 × 262144 × 2688
47	FFN_11	Feed Forward	1 × 262144 × 2688
48	Attention_12	Multi-Head Attention	1 × 262144 × 2688
49	Add_12	Add	1 × 262144 × 2688
50	LayerNorm_12_1	LayerNorm	1 × 262144 × 2688
51	FFN_12	Feed Forward	1 × 262144 × 2688
52	Attention_13	Multi-Head Attention	1 × 262144 × 2688
53	Add_13	Add	1 × 262144 × 2688
54	LayerNorm_13_1	LayerNorm	1 × 262144 × 2688
55	FFN_13	Feed Forward	1 × 262144 × 2688
56	Attention_14	Multi-Head Attention	1 × 262144 × 2688
57	Add_14	Add	1 × 262144 × 2688
58	LayerNorm_14_1	LayerNorm	1 × 262144 × 2688
59	FFN_14	Feed Forward	1 × 262144 × 2688
60	Attention_15	Multi-Head Attention	1 × 262144 × 2688
61	Add_15	Add	1 × 262144 × 2688
62	LayerNorm_15_1	LayerNorm	1 × 262144 × 2688
63	FFN_15	Feed Forward	1 × 262144 × 2688
64	Attention_16	Multi-Head Attention	1 × 262144 × 2688
65	Add_16	Add	1 × 262144 × 2688
66	LayerNorm_16_1	LayerNorm	1 × 262144 × 2688
67	FFN_16	Feed Forward	1 × 262144 × 2688
68	Attention_17	Multi-Head Attention	1 × 262144 × 2688
69	Add_17	Add	1 × 262144 × 2688
70	LayerNorm_17_1	LayerNorm	1 × 262144 × 2688
71	FFN_17	Feed Forward	1 × 262144 × 2688
72	Attention_18	Multi-Head Attention	1 × 262144 × 2688
73	Add_18	Add	1 × 262144 × 2688
74	LayerNorm_18_1	LayerNorm	1 × 262144 × 2688
75	FFN_18	Feed Forward	1 × 262144 × 2688
76	Attention_19	Multi-Head Attention	1 × 262144 × 2688
77	Add_19	Add	1 × 262144 × 2688
78	LayerNorm_19_1	LayerNorm	1 × 262144 × 2688
79	FFN_19	Feed Forward	1 × 262144 × 2688
80	Attention_20	Multi-Head Attention	1 × 262144 × 2688
81	Add_20	Add	1 × 262144 × 2688
82	LayerNorm_20_1	LayerNorm	1 × 262144 × 2688
83	FFN_20	Feed Forward	1 × 262144 × 2688
84	Attention_21	Multi-Head Attention	1 × 262144 × 2688
85	Add_21	Add	1 × 262144 × 2688
86	LayerNorm_21_1	LayerNorm	1 × 262144 × 2688
87	FFN_21	Feed Forward	1 × 262144 × 2688
88	Attention_22	Multi-Head Attention	1 × 262144 × 2688
89	Add_22	Add	1 × 262144 × 2688
90	LayerNorm_22_1	LayerNorm	1 × 262144 × 2688
91	FFN_22	Feed Forward	1 × 262144 × 2688
92	Attention_23	Multi-Head Attention	1 × 262144 × 2688
93	Add_23	Add	1 × 262144 × 2688
94	LayerNorm_23_1	LayerNorm	1 × 262144 × 2688
95	FFN_23	Feed Forward	1 × 262144 × 2688
96	Attention_24	Multi-Head Attention	1 × 262144 × 2688
97	Add_24	Add	1 × 262144 × 2688
98	LayerNorm_24_1	LayerNorm	1 × 262144 × 2688
99	FFN_24	Feed Forward	1 × 262144 × 2688
100	Attention_25	Multi-Head Attention	1 × 262144 × 2688
101	Add_25	Add	1 × 262144 × 2688
102	LayerNorm_25_1	LayerNorm	1 × 262144 × 2688
103	FFN_25	Feed Forward	1 × 262144 × 2688
104	Attention_26	Multi-Head Attention	1 × 262144 × 2688
105	Add_26	Add	1 × 262144 × 2688
106	LayerNorm_26_1	LayerNorm	1 × 262144 × 2688
107	FFN_26	Feed Forward	1 × 262144 × 2688
108	Attention_27	Multi-Head Attention	1 × 262144 × 2688
109	Add_27	Add	1 × 262144 × 2688
110	LayerNorm_27_1	LayerNorm	1 × 262144 × 2688
111	FFN_27	Feed Forward	1 × 262144 × 2688
112	Attention_28	Multi-Head Attention	1 × 262144 × 2688
113	Add_28	Add	1 × 262144 × 2688
114	LayerNorm_28_1	LayerNorm	1 × 262144 × 2688
115	FFN_28	Feed Forward	1 × 262144 × 2688
116	Attention_29	Multi-Head Attention	1 × 262144 × 2688
117	Add_29	Add	1 × 262144 × 2688
118	LayerNorm_29_1	LayerNorm	1 × 262144 × 2688
119	FFN_29	Feed Forward	1 × 262144 × 2688
120	Attention_30	Multi-Head Attention	1 × 262144 × 2688
121	Add_30	Add	1 × 262144 × 2688
122	LayerNorm_30_1	LayerNorm	1 × 262144 × 2688
123	FFN_30	Feed Forward	1 × 262144 × 2688
124	Attention_31	Multi-Head Attention	1 × 262144 × 2688
125	Add_31	Add	1 × 262144 × 2688
126	LayerNorm_31_1	LayerNorm	1 × 262144 × 2688
127	FFN_31	Feed Forward	1 × 262144 × 2688
128	Attention_32	Multi-Head Attention	1 × 262144 × 2688
129	Add_32	Add	1 × 262144 × 2688
130	LayerNorm_32_1	LayerNorm	1 × 262144 × 2688
131	FFN_32	Feed Forward	1 × 262144 × 2688
132	Attention_33	Multi-Head Attention	1 × 262144 × 2688
133	Add_33	Add	1 × 262144 × 2688
134	LayerNorm_33_1	LayerNorm	1 × 262144 × 2688
135	FFN_33	Feed Forward	1 × 262144 × 2688
136	Attention_34	Multi-Head Attention	1 × 262144 × 2688
137	Add_34	Add	1 × 262144 × 2688
138	LayerNorm_34_1	LayerNorm	1 × 262144 × 2688
139	FFN_34	Feed Forward	1 × 262144 × 2688
140	Attention_35	Multi-Head Attention	1 × 262144 × 2688
141	Add_35	Add	1 × 262144 × 2688
142	LayerNorm_35_1	LayerNorm	1 × 262144 × 2688
143	FFN_35	Feed Forward	1 × 262144 × 2688
144	Attention_36	Multi-Head Attention	1 × 262144 × 2688
145	Add_36	Add	1 × 262144 × 2688
146	LayerNorm_36_1	LayerNorm	1 × 262144 × 2688
147	FFN_36	Feed Forward	1 × 262144 × 2688
148	Attention_37	Multi-Head Attention	1 × 262144 × 2688
149	Add_37	Add	1 × 262144 × 2688
150	LayerNorm_37_1	LayerNorm	1 × 262144 × 2688
151	FFN_37	Feed Forward	1 × 262144 × 2688
152	Attention_38	Multi-Head Attention	1 × 262144 × 2688
153	Add_38	Add	1 × 262144 × 2688
154	LayerNorm_38_1	LayerNorm	1 × 262144 × 2688
155	FFN_38	Feed Forward	1 × 262144 × 2688
156	Attention_39	Multi-Head Attention	1 × 262144 × 2688
157	Add_39	Add	1 × 262144 × 2688
158	LayerNorm_39_1	LayerNorm	1 × 262144 × 2688
159	FFN_39	Feed Forward	1 × 262144 × 2688
160	Attention_40	Multi-Head Attention	1 × 262144 × 2688
161	Add_40	Add	1 × 262144 × 2688
162	LayerNorm_40_1	LayerNorm	1 × 262144 × 2688
163	FFN_40	Feed Forward	1 × 262144 × 2688
164	Attention_41	Multi-Head Attention	1 × 262144 × 2688
165	Add_41	Add	1 × 262144 × 2688
166	LayerNorm_41_1	LayerNorm	1 × 262144 × 2688
167	FFN_41	Feed Forward	1 × 262144 × 2688
168	Attention_42	Multi-Head Attention	1 × 262144 × 2688
169	Add_42	Add	1 × 262144 × 2688
170	LayerNorm_42_1	LayerNorm	1 × 262144 × 2688
171	FFN_42	Feed Forward	1 × 262144 × 2688
172	Attention_43	Multi-Head Attention	1 × 262144 × 2688
173	Add_43	Add	1 × 262144 × 2688
174	LayerNorm_43_1	LayerNorm	1 × 262144 × 2688
175	FFN_43	Feed Forward	1 × 262144 × 2688
176	Attention_44	Multi-Head Attention	1 × 262144 × 2688
177	Add_44	Add	1 × 262144 × 2688
178	LayerNorm_44_1	LayerNorm	1 × 262144 × 2688
179	FFN_44	Feed Forward	1 × 262144 × 2688
180	Attention_45	Multi-Head Attention	1 × 262144 × 2688
181	Add_45	Add	1 × 262144 × 2688
182	LayerNorm_45_1	LayerNorm	1 × 262144 × 2688
183	FFN_45	Feed Forward	1 × 262144 × 2688
184	Attention_46	Multi-Head Attention	1 × 262144 × 2688
185	Add_46	Add	1 × 262144 × 2688
186	LayerNorm_46_1	LayerNorm	1 × 262144 × 2688
187	FFN_46	Feed Forward	1 × 262144 × 2688
188	Attention_47	Multi-Head Attention	1 × 262144 × 2688
189	Add_47	Add	1 × 262144 × 2688
190	LayerNorm_47_1	LayerNorm	1 × 262144 × 2688
191	FFN_47	Feed Forward	1 × 262144 × 2688
192	Attention_48	Multi-Head Attention	1 × 262144 × 2688
193	Add_48	Add	1 × 262144 × 2688
194	LayerNorm_48_1	LayerNorm	1 × 262144 × 2688
195	FFN_48	Feed Forward	1 × 262144 × 2688
196	Attention_49	Multi-Head Attention	1 × 262144 × 2688
197	Add_49	Add	1 × 262144 × 2688
198	LayerNorm_49_1	LayerNorm	1 × 262144 × 2688
199	FFN_49	Feed Forward	1 × 262144 × 2688
200	Attention_50	Multi-Head Attention	1 × 262144 × 2688
201	Add_50	Add	1 × 262144 × 2688
202	LayerNorm_50_1	LayerNorm	1 × 262144 × 2688
203	FFN_50	Feed Forward	1 × 262144 × 2688
204	Attention_51	Multi-Head Attention	1 × 262144 × 2688
205	Add_51	Add	1 × 262144 × 2688
206	LayerNorm_51_1	LayerNorm	1 × 262144 × 2688
207	FFN_51	Feed Forward	1 × 262144 × 2688
208	Attention_52	Multi-Head Attention	1 × 262144 × 2688
209	Add_52	Add	1 × 262144 × 2688
210	LayerNorm_52_1	LayerNorm	1 × 262144 × 2688
211	FFN_52	Feed Forward	1 × 262144 × 2688
212	Output	Output	1 × 262144 × 2688

What the verifier says

info52 attention layers at embedDim 2688 cache full per-head K/V: about 546 KB per token at fp16, which dominates memory at long context. Grouped-query attention (e.g. 8:1) would cut this ~8×; multi-head latent attention (MLA) shrinks it ~10× or more. This is the move production LLMs make; it does not change the parameter count. Fix: Switch attention to groupedQueryAttention (set numKVHeads below numHeads, e.g. numHeads/4) or mla (a low-rank cached latent).
full-mha-serving-cost

infoAt 52 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

warnAcross 52 attention layers this design caches 546 KB per token, so a single 8,192-token sequence needs ~4.6 GB of KV cache before weights or activations. That exceeds the 4 GB budget this rule assumes for serving headroom. Fix: Cut KV width: raise the GQA ratio (fewer numKVHeads), switch to MLA, reduce depth or embedDim, or accept a shorter serving context.
kv-cache-context-budget

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/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4 --plan --share

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

Other NemotronH_Nano_Omni_Reasoning_V3 checkpoints

Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16

3.08B derived · -90.7% against the checkpoint

Nemotron-3-Nano-Omni-30B-A3B-Reasoning-FP8

3.08B derived · -90.7% against the checkpoint