Models / nemotron_h

NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4: GPU memory to fine-tune

Fine-tuning NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 at batch 1 and 2048 tokens takes about 2.39 TB of GPU memory for a full fine-tune (at least 20 x H200 sharded), and 264 GB with LoRA r=16 on a bf16 base (at least 3 x H200 sharded). Estimated from the model's structure, not measured. What fits on which card.

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

124B

123,611,147,264 parameters

In the published checkpoint

67.23B

67,228,556,288 scalars · safetensors.total, read 2026-08-24

Delta

+83.9%

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, 2048 tokens over 50,000 samples (assumed; no dataset attached).

Layers

278

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$2804.60

Memory: what fits on which card

Largest batch per card. Training is sized at 2048 tokens per sample, serving at 4096 tokens per sequence.

Full fine-tune, batch 1 at 2048 tokens

2.39 TB

LoRA r=16, batch 1 at 2048 tokens

264 GB

Serve: bf16 weights

247 GB

Serve: KV cache, 4096 tokens

34 MB

Serve: recurrent state

86 MB

Card	Full fine-tune	LoRA r=16	Serve bf16, 4096 tokens
T4 16 GB	does not fit at batch 1	does not fit at batch 1	does not fit at batch 1
L4 24 GB	does not fit at batch 1	does not fit at batch 1	does not fit at batch 1
A100 40 GB	does not fit at batch 1	does not fit at batch 1	does not fit at batch 1
A100 / H100 80 GB	does not fit at batch 1	does not fit at batch 1	does not fit at batch 1
H200 141 GB	does not fit at batch 1	does not fit at batch 1	does not fit at batch 1
Sharded, by memory alone	at least 20 x H200 or 35 x 80 GB	at least 3 x H200 or 4 x 80 GB	at least 3 x H200 or 4 x 80 GB

Training: 2048 tokens per sample (2048, or the model's context when shorter; the structure table is at the config's maximum, 262144), AdamW with bf16 autocast, SDPA attention (no score matrix stored), 1.15x headroom on every card. Full fine-tune holds weights, gradients and optimizer state (1.2 x 16 bytes per parameter) plus activations. LoRA is r=16 on every linear layer but the head, base frozen in bf16 (2 bytes per parameter), adapters at full AdamW state, same activations. A two-card split is a contiguous layer cut (pipeline parallel), computed only where one card cannot hold batch 1. "Sharded" is a floor: the batch-1 peak with headroom divided across cards as if it split perfectly evenly (FSDP / ZeRO-3 for training, tensor parallel for serving), with no communication buffers counted. Serving: bf16 weights plus the KV cache of 4096 tokens per sequence plus the fixed recurrent state of the SSM / linear-attention layers; prefill working set, CUDA context and framework reserve not counted.

These are estimates from the graph, not measurements. Against measured runs, held out by architecture, the estimate flagged 113 of the 131 jobs that did not fit their card and said "fits" on the other 18. Every measured run.

Structure

281 nodes. Output shapes are propagated from the input shape, batch dimension excluded, at 262144 tokens: the config's maximum context, not the 2048 the memory above is sized at.

	Layer	Type	Output shape
1	Input	Input	1 × 262144
2	Embedding	Embedding	1 × 262144 × 4096
3	RMSNorm_1_1	RMSNorm	1 × 262144 × 4096
4	Mamba_1	Mamba-2 (SSD)	1 × 262144 × 4096
5	Add_1_attn	Add	1 × 262144 × 4096
6	RMSNorm_2_2	RMSNorm	1 × 262144 × 4096
7	MoE_2	Shared-Expert MoE	1 × 262144 × 4096
8	Add_2_ffn	Add	1 × 262144 × 4096
9	RMSNorm_3_1	RMSNorm	1 × 262144 × 4096
10	Mamba_3	Mamba-2 (SSD)	1 × 262144 × 4096
11	Add_3_attn	Add	1 × 262144 × 4096
12	RMSNorm_4_2	RMSNorm	1 × 262144 × 4096
13	MoE_4	Shared-Expert MoE	1 × 262144 × 4096
14	Add_4_ffn	Add	1 × 262144 × 4096
15	RMSNorm_5_1	RMSNorm	1 × 262144 × 4096
16	Mamba_5	Mamba-2 (SSD)	1 × 262144 × 4096
17	Add_5_attn	Add	1 × 262144 × 4096
18	RMSNorm_6_2	RMSNorm	1 × 262144 × 4096
19	MoE_6	Shared-Expert MoE	1 × 262144 × 4096
20	Add_6_ffn	Add	1 × 262144 × 4096
21	RMSNorm_7_1	RMSNorm	1 × 262144 × 4096
22	Mamba_7	Mamba-2 (SSD)	1 × 262144 × 4096
23	Add_7_attn	Add	1 × 262144 × 4096
24	RMSNorm_8_1	RMSNorm	1 × 262144 × 4096
25	Attention_8	Grouped Query Attn	1 × 262144 × 4096
26	Add_8_attn	Add	1 × 262144 × 4096
27	RMSNorm_9_2	RMSNorm	1 × 262144 × 4096
28	MoE_9	Shared-Expert MoE	1 × 262144 × 4096
29	Add_9_ffn	Add	1 × 262144 × 4096
30	RMSNorm_10_1	RMSNorm	1 × 262144 × 4096
31	Mamba_10	Mamba-2 (SSD)	1 × 262144 × 4096
32	Add_10_attn	Add	1 × 262144 × 4096
33	RMSNorm_11_2	RMSNorm	1 × 262144 × 4096
34	MoE_11	Shared-Expert MoE	1 × 262144 × 4096
35	Add_11_ffn	Add	1 × 262144 × 4096
36	RMSNorm_12_1	RMSNorm	1 × 262144 × 4096
37	Mamba_12	Mamba-2 (SSD)	1 × 262144 × 4096
38	Add_12_attn	Add	1 × 262144 × 4096
39	RMSNorm_13_2	RMSNorm	1 × 262144 × 4096
40	MoE_13	Shared-Expert MoE	1 × 262144 × 4096
41	Add_13_ffn	Add	1 × 262144 × 4096
42	RMSNorm_14_1	RMSNorm	1 × 262144 × 4096
43	Mamba_14	Mamba-2 (SSD)	1 × 262144 × 4096
44	Add_14_attn	Add	1 × 262144 × 4096
45	RMSNorm_15_2	RMSNorm	1 × 262144 × 4096
46	MoE_15	Shared-Expert MoE	1 × 262144 × 4096
47	Add_15_ffn	Add	1 × 262144 × 4096
48	RMSNorm_16_1	RMSNorm	1 × 262144 × 4096
49	Mamba_16	Mamba-2 (SSD)	1 × 262144 × 4096
50	Add_16_attn	Add	1 × 262144 × 4096
51	RMSNorm_17_1	RMSNorm	1 × 262144 × 4096
52	Attention_17	Grouped Query Attn	1 × 262144 × 4096
53	Add_17_attn	Add	1 × 262144 × 4096
54	RMSNorm_18_2	RMSNorm	1 × 262144 × 4096
55	MoE_18	Shared-Expert MoE	1 × 262144 × 4096
56	Add_18_ffn	Add	1 × 262144 × 4096
57	RMSNorm_19_1	RMSNorm	1 × 262144 × 4096
58	Mamba_19	Mamba-2 (SSD)	1 × 262144 × 4096
59	Add_19_attn	Add	1 × 262144 × 4096
60	RMSNorm_20_2	RMSNorm	1 × 262144 × 4096
61	MoE_20	Shared-Expert MoE	1 × 262144 × 4096
62	Add_20_ffn	Add	1 × 262144 × 4096
63	RMSNorm_21_1	RMSNorm	1 × 262144 × 4096
64	Mamba_21	Mamba-2 (SSD)	1 × 262144 × 4096
65	Add_21_attn	Add	1 × 262144 × 4096
66	RMSNorm_22_2	RMSNorm	1 × 262144 × 4096
67	MoE_22	Shared-Expert MoE	1 × 262144 × 4096
68	Add_22_ffn	Add	1 × 262144 × 4096
69	RMSNorm_23_1	RMSNorm	1 × 262144 × 4096
70	Mamba_23	Mamba-2 (SSD)	1 × 262144 × 4096
71	Add_23_attn	Add	1 × 262144 × 4096
72	RMSNorm_24_2	RMSNorm	1 × 262144 × 4096
73	MoE_24	Shared-Expert MoE	1 × 262144 × 4096
74	Add_24_ffn	Add	1 × 262144 × 4096
75	RMSNorm_25_1	RMSNorm	1 × 262144 × 4096
76	Mamba_25	Mamba-2 (SSD)	1 × 262144 × 4096
77	Add_25_attn	Add	1 × 262144 × 4096
78	RMSNorm_26_1	RMSNorm	1 × 262144 × 4096
79	Attention_26	Grouped Query Attn	1 × 262144 × 4096
80	Add_26_attn	Add	1 × 262144 × 4096
81	RMSNorm_27_2	RMSNorm	1 × 262144 × 4096
82	MoE_27	Shared-Expert MoE	1 × 262144 × 4096
83	Add_27_ffn	Add	1 × 262144 × 4096
84	RMSNorm_28_1	RMSNorm	1 × 262144 × 4096
85	Mamba_28	Mamba-2 (SSD)	1 × 262144 × 4096
86	Add_28_attn	Add	1 × 262144 × 4096
87	RMSNorm_29_2	RMSNorm	1 × 262144 × 4096
88	MoE_29	Shared-Expert MoE	1 × 262144 × 4096
89	Add_29_ffn	Add	1 × 262144 × 4096
90	RMSNorm_30_1	RMSNorm	1 × 262144 × 4096
91	Mamba_30	Mamba-2 (SSD)	1 × 262144 × 4096
92	Add_30_attn	Add	1 × 262144 × 4096
93	RMSNorm_31_2	RMSNorm	1 × 262144 × 4096
94	MoE_31	Shared-Expert MoE	1 × 262144 × 4096
95	Add_31_ffn	Add	1 × 262144 × 4096
96	RMSNorm_32_1	RMSNorm	1 × 262144 × 4096
97	Mamba_32	Mamba-2 (SSD)	1 × 262144 × 4096
98	Add_32_attn	Add	1 × 262144 × 4096
99	RMSNorm_33_2	RMSNorm	1 × 262144 × 4096
100	MoE_33	Shared-Expert MoE	1 × 262144 × 4096
101	Add_33_ffn	Add	1 × 262144 × 4096
102	RMSNorm_34_1	RMSNorm	1 × 262144 × 4096
103	Mamba_34	Mamba-2 (SSD)	1 × 262144 × 4096
104	Add_34_attn	Add	1 × 262144 × 4096
105	RMSNorm_35_2	RMSNorm	1 × 262144 × 4096
106	MoE_35	Shared-Expert MoE	1 × 262144 × 4096
107	Add_35_ffn	Add	1 × 262144 × 4096
108	RMSNorm_36_1	RMSNorm	1 × 262144 × 4096
109	Mamba_36	Mamba-2 (SSD)	1 × 262144 × 4096
110	Add_36_attn	Add	1 × 262144 × 4096
111	RMSNorm_37_1	RMSNorm	1 × 262144 × 4096
112	Attention_37	Grouped Query Attn	1 × 262144 × 4096
113	Add_37_attn	Add	1 × 262144 × 4096
114	RMSNorm_38_2	RMSNorm	1 × 262144 × 4096
115	MoE_38	Shared-Expert MoE	1 × 262144 × 4096
116	Add_38_ffn	Add	1 × 262144 × 4096
117	RMSNorm_39_1	RMSNorm	1 × 262144 × 4096
118	Mamba_39	Mamba-2 (SSD)	1 × 262144 × 4096
119	Add_39_attn	Add	1 × 262144 × 4096
120	RMSNorm_40_2	RMSNorm	1 × 262144 × 4096
121	MoE_40	Shared-Expert MoE	1 × 262144 × 4096
122	Add_40_ffn	Add	1 × 262144 × 4096
123	RMSNorm_41_1	RMSNorm	1 × 262144 × 4096
124	Mamba_41	Mamba-2 (SSD)	1 × 262144 × 4096
125	Add_41_attn	Add	1 × 262144 × 4096
126	RMSNorm_42_2	RMSNorm	1 × 262144 × 4096
127	MoE_42	Shared-Expert MoE	1 × 262144 × 4096
128	Add_42_ffn	Add	1 × 262144 × 4096
129	RMSNorm_43_1	RMSNorm	1 × 262144 × 4096
130	Mamba_43	Mamba-2 (SSD)	1 × 262144 × 4096
131	Add_43_attn	Add	1 × 262144 × 4096
132	RMSNorm_44_2	RMSNorm	1 × 262144 × 4096
133	MoE_44	Shared-Expert MoE	1 × 262144 × 4096
134	Add_44_ffn	Add	1 × 262144 × 4096
135	RMSNorm_45_1	RMSNorm	1 × 262144 × 4096
136	Mamba_45	Mamba-2 (SSD)	1 × 262144 × 4096
137	Add_45_attn	Add	1 × 262144 × 4096
138	RMSNorm_46_2	RMSNorm	1 × 262144 × 4096
139	MoE_46	Shared-Expert MoE	1 × 262144 × 4096
140	Add_46_ffn	Add	1 × 262144 × 4096
141	RMSNorm_47_1	RMSNorm	1 × 262144 × 4096
142	Mamba_47	Mamba-2 (SSD)	1 × 262144 × 4096
143	Add_47_attn	Add	1 × 262144 × 4096
144	RMSNorm_48_1	RMSNorm	1 × 262144 × 4096
145	Attention_48	Grouped Query Attn	1 × 262144 × 4096
146	Add_48_attn	Add	1 × 262144 × 4096
147	RMSNorm_49_2	RMSNorm	1 × 262144 × 4096
148	MoE_49	Shared-Expert MoE	1 × 262144 × 4096
149	Add_49_ffn	Add	1 × 262144 × 4096
150	RMSNorm_50_1	RMSNorm	1 × 262144 × 4096
151	Mamba_50	Mamba-2 (SSD)	1 × 262144 × 4096
152	Add_50_attn	Add	1 × 262144 × 4096
153	RMSNorm_51_2	RMSNorm	1 × 262144 × 4096
154	MoE_51	Shared-Expert MoE	1 × 262144 × 4096
155	Add_51_ffn	Add	1 × 262144 × 4096
156	RMSNorm_52_1	RMSNorm	1 × 262144 × 4096
157	Mamba_52	Mamba-2 (SSD)	1 × 262144 × 4096
158	Add_52_attn	Add	1 × 262144 × 4096
159	RMSNorm_53_2	RMSNorm	1 × 262144 × 4096
160	MoE_53	Shared-Expert MoE	1 × 262144 × 4096
161	Add_53_ffn	Add	1 × 262144 × 4096
162	RMSNorm_54_1	RMSNorm	1 × 262144 × 4096
163	Mamba_54	Mamba-2 (SSD)	1 × 262144 × 4096
164	Add_54_attn	Add	1 × 262144 × 4096
165	RMSNorm_55_2	RMSNorm	1 × 262144 × 4096
166	MoE_55	Shared-Expert MoE	1 × 262144 × 4096
167	Add_55_ffn	Add	1 × 262144 × 4096
168	RMSNorm_56_1	RMSNorm	1 × 262144 × 4096
169	Mamba_56	Mamba-2 (SSD)	1 × 262144 × 4096
170	Add_56_attn	Add	1 × 262144 × 4096
171	RMSNorm_57_2	RMSNorm	1 × 262144 × 4096
172	MoE_57	Shared-Expert MoE	1 × 262144 × 4096
173	Add_57_ffn	Add	1 × 262144 × 4096
174	RMSNorm_58_1	RMSNorm	1 × 262144 × 4096
175	Mamba_58	Mamba-2 (SSD)	1 × 262144 × 4096
176	Add_58_attn	Add	1 × 262144 × 4096
177	RMSNorm_59_1	RMSNorm	1 × 262144 × 4096
178	Attention_59	Grouped Query Attn	1 × 262144 × 4096
179	Add_59_attn	Add	1 × 262144 × 4096
180	RMSNorm_60_2	RMSNorm	1 × 262144 × 4096
181	MoE_60	Shared-Expert MoE	1 × 262144 × 4096
182	Add_60_ffn	Add	1 × 262144 × 4096
183	RMSNorm_61_1	RMSNorm	1 × 262144 × 4096
184	Mamba_61	Mamba-2 (SSD)	1 × 262144 × 4096
185	Add_61_attn	Add	1 × 262144 × 4096
186	RMSNorm_62_2	RMSNorm	1 × 262144 × 4096
187	MoE_62	Shared-Expert MoE	1 × 262144 × 4096
188	Add_62_ffn	Add	1 × 262144 × 4096
189	RMSNorm_63_1	RMSNorm	1 × 262144 × 4096
190	Mamba_63	Mamba-2 (SSD)	1 × 262144 × 4096
191	Add_63_attn	Add	1 × 262144 × 4096
192	RMSNorm_64_2	RMSNorm	1 × 262144 × 4096
193	MoE_64	Shared-Expert MoE	1 × 262144 × 4096
194	Add_64_ffn	Add	1 × 262144 × 4096
195	RMSNorm_65_1	RMSNorm	1 × 262144 × 4096
196	Mamba_65	Mamba-2 (SSD)	1 × 262144 × 4096
197	Add_65_attn	Add	1 × 262144 × 4096
198	RMSNorm_66_2	RMSNorm	1 × 262144 × 4096
199	MoE_66	Shared-Expert MoE	1 × 262144 × 4096
200	Add_66_ffn	Add	1 × 262144 × 4096
201	RMSNorm_67_1	RMSNorm	1 × 262144 × 4096
202	Mamba_67	Mamba-2 (SSD)	1 × 262144 × 4096
203	Add_67_attn	Add	1 × 262144 × 4096
204	RMSNorm_68_2	RMSNorm	1 × 262144 × 4096
205	MoE_68	Shared-Expert MoE	1 × 262144 × 4096
206	Add_68_ffn	Add	1 × 262144 × 4096
207	RMSNorm_69_1	RMSNorm	1 × 262144 × 4096
208	Mamba_69	Mamba-2 (SSD)	1 × 262144 × 4096
209	Add_69_attn	Add	1 × 262144 × 4096
210	RMSNorm_70_1	RMSNorm	1 × 262144 × 4096
211	Attention_70	Grouped Query Attn	1 × 262144 × 4096
212	Add_70_attn	Add	1 × 262144 × 4096
213	RMSNorm_71_2	RMSNorm	1 × 262144 × 4096
214	MoE_71	Shared-Expert MoE	1 × 262144 × 4096
215	Add_71_ffn	Add	1 × 262144 × 4096
216	RMSNorm_72_1	RMSNorm	1 × 262144 × 4096
217	Mamba_72	Mamba-2 (SSD)	1 × 262144 × 4096
218	Add_72_attn	Add	1 × 262144 × 4096
219	RMSNorm_73_2	RMSNorm	1 × 262144 × 4096
220	MoE_73	Shared-Expert MoE	1 × 262144 × 4096
221	Add_73_ffn	Add	1 × 262144 × 4096
222	RMSNorm_74_1	RMSNorm	1 × 262144 × 4096
223	Mamba_74	Mamba-2 (SSD)	1 × 262144 × 4096
224	Add_74_attn	Add	1 × 262144 × 4096
225	RMSNorm_75_2	RMSNorm	1 × 262144 × 4096
226	MoE_75	Shared-Expert MoE	1 × 262144 × 4096
227	Add_75_ffn	Add	1 × 262144 × 4096
228	RMSNorm_76_1	RMSNorm	1 × 262144 × 4096
229	Mamba_76	Mamba-2 (SSD)	1 × 262144 × 4096
230	Add_76_attn	Add	1 × 262144 × 4096
231	RMSNorm_77_2	RMSNorm	1 × 262144 × 4096
232	MoE_77	Shared-Expert MoE	1 × 262144 × 4096
233	Add_77_ffn	Add	1 × 262144 × 4096
234	RMSNorm_78_1	RMSNorm	1 × 262144 × 4096
235	Mamba_78	Mamba-2 (SSD)	1 × 262144 × 4096
236	Add_78_attn	Add	1 × 262144 × 4096
237	RMSNorm_79_1	RMSNorm	1 × 262144 × 4096
238	Attention_79	Grouped Query Attn	1 × 262144 × 4096
239	Add_79_attn	Add	1 × 262144 × 4096
240	RMSNorm_80_2	RMSNorm	1 × 262144 × 4096
241	MoE_80	Shared-Expert MoE	1 × 262144 × 4096
242	Add_80_ffn	Add	1 × 262144 × 4096
243	RMSNorm_81_1	RMSNorm	1 × 262144 × 4096
244	Mamba_81	Mamba-2 (SSD)	1 × 262144 × 4096
245	Add_81_attn	Add	1 × 262144 × 4096
246	RMSNorm_82_2	RMSNorm	1 × 262144 × 4096
247	MoE_82	Shared-Expert MoE	1 × 262144 × 4096
248	Add_82_ffn	Add	1 × 262144 × 4096
249	RMSNorm_83_1	RMSNorm	1 × 262144 × 4096
250	Mamba_83	Mamba-2 (SSD)	1 × 262144 × 4096
251	Add_83_attn	Add	1 × 262144 × 4096
252	RMSNorm_84_2	RMSNorm	1 × 262144 × 4096
253	MoE_84	Shared-Expert MoE	1 × 262144 × 4096
254	Add_84_ffn	Add	1 × 262144 × 4096
255	RMSNorm_85_1	RMSNorm	1 × 262144 × 4096
256	Mamba_85	Mamba-2 (SSD)	1 × 262144 × 4096
257	Add_85_attn	Add	1 × 262144 × 4096
258	RMSNorm_86_2	RMSNorm	1 × 262144 × 4096
259	MoE_86	Shared-Expert MoE	1 × 262144 × 4096
260	Add_86_ffn	Add	1 × 262144 × 4096
261	RMSNorm_87_1	RMSNorm	1 × 262144 × 4096
262	Mamba_87	Mamba-2 (SSD)	1 × 262144 × 4096
263	Add_87_attn	Add	1 × 262144 × 4096
264	RMSNorm_88_2	RMSNorm	1 × 262144 × 4096
265	MoE_88	Shared-Expert MoE	1 × 262144 × 4096
266	Add_88_ffn	Add	1 × 262144 × 4096
267	MTP_1_HiddenNorm	RMSNorm	1 × 262144 × 4096
268	MTP_1_EmbeddingNorm	RMSNorm	1 × 262144 × 4096
269	MTP_1_Concat	Concatenate	1 × 262144 × 8192
270	MTP_1_Projection	Linear	1 × 262144 × 4096
271	RMSNorm_MTP_1_1	RMSNorm	1 × 262144 × 4096
272	Attention_MTP_1	Grouped Query Attn	1 × 262144 × 4096
273	Add_MTP_1_attn	Add	1 × 262144 × 4096
274	RMSNorm_MTP_1_2	RMSNorm	1 × 262144 × 4096
275	MoE_MTP_1	Shared-Expert MoE	1 × 262144 × 4096
276	Add_MTP_1_ffn	Add	1 × 262144 × 4096
277	MTP_1_Norm	RMSNorm	1 × 262144 × 4096
278	Final_RMSNorm	RMSNorm	1 × 262144 × 4096
279	LM_Head	Linear	1 × 262144 × 131072
280	Output	Output	1 × 262144 × 131072
281	MTP_1_Output	Output	1 × 262144 × 4096

What the verifier says

warn9 attention layer(s) present but no positional encoding found. Attention is permutation-invariant, without position information the model cannot distinguish token order. Fix: Add a PositionalEncoding (sinusoidal) or RoPE layer before the first attention layer.
attention-no-pe

infoAt 9 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

What changed upstream

The change tracker follows nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4: every revision it has read, what changed in the config, tokenizer, chat template or weights between them, and an OSV advisory per change.

Revision history and changes Badge JSON OSV advisories All tracked changes

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

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

Other nemotron_h checkpoints

NVIDIA-Nemotron-3-Nano-30B-A3B-BF16

31.58B derived · +0.00% against the checkpoint

NVIDIA-Nemotron-3-Nano-30B-A3B-FP8

31.58B derived · +0.00% against the checkpoint

NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4

31.58B derived · +73.1% against the checkpoint

NVIDIA-Nemotron-3-Super-120B-A12B-BF16

124B derived · +0.00% against the checkpoint