N Neurarch Architectures Models Checks Data Docs Open the app

Models / nemotron-nas

Llama-3_3-Nemotron-Super-49B-v1

Reconstructed from its own config.json with no weights read. 134K 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
29.62B
29,617,427,712 parameters
In the published checkpoint
49.87B
49,867,145,216 scalars · safetensors.total, read 2025-10-15
Delta
-40.6%

custom-code This repository ships its own modeling code (`auto_map`, e.g. `configuration_decilm.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
484
Will it forward-pass
Yes
Priced on
A10G (24GB)
Est. one run
$561534.71
CardMemory
T4 (16GB)weights + activationsdoes not fit
A100 (40GB)weights + activationsdoes not fit
H100 (80GB)weights + activationsdoes not fit

Structure

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

LayerTypeOutput shape
1InputInput1 × 131072
2EmbeddingEmbedding1 × 131072 × 8192
3RoPERoPE1 × 131072 × 8192
4RMSNorm_1_1RMSNorm1 × 131072 × 8192
5Attention_1Grouped Query Attn1 × 131072 × 8192
6Add_1_attnAdd1 × 131072 × 8192
7RMSNorm_1_2RMSNorm1 × 131072 × 8192
8FFN_1SwiGLU1 × 131072 × 8192
9Add_1_ffnAdd1 × 131072 × 8192
10RMSNorm_2_1RMSNorm1 × 131072 × 8192
11Attention_2Grouped Query Attn1 × 131072 × 8192
12Add_2_attnAdd1 × 131072 × 8192
13RMSNorm_2_2RMSNorm1 × 131072 × 8192
14FFN_2SwiGLU1 × 131072 × 8192
15Add_2_ffnAdd1 × 131072 × 8192
16RMSNorm_3_1RMSNorm1 × 131072 × 8192
17Attention_3Grouped Query Attn1 × 131072 × 8192
18Add_3_attnAdd1 × 131072 × 8192
19RMSNorm_3_2RMSNorm1 × 131072 × 8192
20FFN_3SwiGLU1 × 131072 × 8192
21Add_3_ffnAdd1 × 131072 × 8192
22RMSNorm_4_1RMSNorm1 × 131072 × 8192
23Attention_4Grouped Query Attn1 × 131072 × 8192
24Add_4_attnAdd1 × 131072 × 8192
25RMSNorm_4_2RMSNorm1 × 131072 × 8192
26FFN_4SwiGLU1 × 131072 × 8192
27Add_4_ffnAdd1 × 131072 × 8192
28RMSNorm_5_1RMSNorm1 × 131072 × 8192
29Attention_5Grouped Query Attn1 × 131072 × 8192
30Add_5_attnAdd1 × 131072 × 8192
31RMSNorm_5_2RMSNorm1 × 131072 × 8192
32FFN_5SwiGLU1 × 131072 × 8192
33Add_5_ffnAdd1 × 131072 × 8192
34RMSNorm_6_1RMSNorm1 × 131072 × 8192
35Attention_6Grouped Query Attn1 × 131072 × 8192
36Add_6_attnAdd1 × 131072 × 8192
37RMSNorm_6_2RMSNorm1 × 131072 × 8192
38FFN_6SwiGLU1 × 131072 × 8192
39Add_6_ffnAdd1 × 131072 × 8192
40RMSNorm_7_1RMSNorm1 × 131072 × 8192
41Attention_7Grouped Query Attn1 × 131072 × 8192
42Add_7_attnAdd1 × 131072 × 8192
43RMSNorm_7_2RMSNorm1 × 131072 × 8192
44FFN_7SwiGLU1 × 131072 × 8192
45Add_7_ffnAdd1 × 131072 × 8192
46RMSNorm_8_1RMSNorm1 × 131072 × 8192
47Attention_8Grouped Query Attn1 × 131072 × 8192
48Add_8_attnAdd1 × 131072 × 8192
49RMSNorm_8_2RMSNorm1 × 131072 × 8192
50FFN_8SwiGLU1 × 131072 × 8192
51Add_8_ffnAdd1 × 131072 × 8192
52RMSNorm_9_1RMSNorm1 × 131072 × 8192
53Attention_9Grouped Query Attn1 × 131072 × 8192
54Add_9_attnAdd1 × 131072 × 8192
55RMSNorm_9_2RMSNorm1 × 131072 × 8192
56FFN_9SwiGLU1 × 131072 × 8192
57Add_9_ffnAdd1 × 131072 × 8192
58RMSNorm_10_1RMSNorm1 × 131072 × 8192
59Attention_10Grouped Query Attn1 × 131072 × 8192
60Add_10_attnAdd1 × 131072 × 8192
61RMSNorm_10_2RMSNorm1 × 131072 × 8192
62FFN_10SwiGLU1 × 131072 × 8192
63Add_10_ffnAdd1 × 131072 × 8192
64RMSNorm_11_1RMSNorm1 × 131072 × 8192
65Attention_11Grouped Query Attn1 × 131072 × 8192
66Add_11_attnAdd1 × 131072 × 8192
67RMSNorm_11_2RMSNorm1 × 131072 × 8192
68FFN_11SwiGLU1 × 131072 × 8192
69Add_11_ffnAdd1 × 131072 × 8192
70RMSNorm_12_1RMSNorm1 × 131072 × 8192
71Attention_12Grouped Query Attn1 × 131072 × 8192
72Add_12_attnAdd1 × 131072 × 8192
73RMSNorm_12_2RMSNorm1 × 131072 × 8192
74FFN_12SwiGLU1 × 131072 × 8192
75Add_12_ffnAdd1 × 131072 × 8192
76RMSNorm_13_1RMSNorm1 × 131072 × 8192
77Attention_13Grouped Query Attn1 × 131072 × 8192
78Add_13_attnAdd1 × 131072 × 8192
79RMSNorm_13_2RMSNorm1 × 131072 × 8192
80FFN_13SwiGLU1 × 131072 × 8192
81Add_13_ffnAdd1 × 131072 × 8192
82RMSNorm_14_1RMSNorm1 × 131072 × 8192
83Attention_14Grouped Query Attn1 × 131072 × 8192
84Add_14_attnAdd1 × 131072 × 8192
85RMSNorm_14_2RMSNorm1 × 131072 × 8192
86FFN_14SwiGLU1 × 131072 × 8192
87Add_14_ffnAdd1 × 131072 × 8192
88RMSNorm_15_1RMSNorm1 × 131072 × 8192
89Attention_15Grouped Query Attn1 × 131072 × 8192
90Add_15_attnAdd1 × 131072 × 8192
91RMSNorm_15_2RMSNorm1 × 131072 × 8192
92FFN_15SwiGLU1 × 131072 × 8192
93Add_15_ffnAdd1 × 131072 × 8192
94RMSNorm_16_1RMSNorm1 × 131072 × 8192
95Attention_16Grouped Query Attn1 × 131072 × 8192
96Add_16_attnAdd1 × 131072 × 8192
97RMSNorm_16_2RMSNorm1 × 131072 × 8192
98FFN_16SwiGLU1 × 131072 × 8192
99Add_16_ffnAdd1 × 131072 × 8192
100RMSNorm_17_1RMSNorm1 × 131072 × 8192
101Attention_17Grouped Query Attn1 × 131072 × 8192
102Add_17_attnAdd1 × 131072 × 8192
103RMSNorm_17_2RMSNorm1 × 131072 × 8192
104FFN_17SwiGLU1 × 131072 × 8192
105Add_17_ffnAdd1 × 131072 × 8192
106RMSNorm_18_1RMSNorm1 × 131072 × 8192
107Attention_18Grouped Query Attn1 × 131072 × 8192
108Add_18_attnAdd1 × 131072 × 8192
109RMSNorm_18_2RMSNorm1 × 131072 × 8192
110FFN_18SwiGLU1 × 131072 × 8192
111Add_18_ffnAdd1 × 131072 × 8192
112RMSNorm_19_1RMSNorm1 × 131072 × 8192
113Attention_19Grouped Query Attn1 × 131072 × 8192
114Add_19_attnAdd1 × 131072 × 8192
115RMSNorm_19_2RMSNorm1 × 131072 × 8192
116FFN_19SwiGLU1 × 131072 × 8192
117Add_19_ffnAdd1 × 131072 × 8192
118RMSNorm_20_1RMSNorm1 × 131072 × 8192
119Attention_20Grouped Query Attn1 × 131072 × 8192
120Add_20_attnAdd1 × 131072 × 8192
121RMSNorm_20_2RMSNorm1 × 131072 × 8192
122FFN_20SwiGLU1 × 131072 × 8192
123Add_20_ffnAdd1 × 131072 × 8192
124RMSNorm_21_1RMSNorm1 × 131072 × 8192
125Attention_21Grouped Query Attn1 × 131072 × 8192
126Add_21_attnAdd1 × 131072 × 8192
127RMSNorm_21_2RMSNorm1 × 131072 × 8192
128FFN_21SwiGLU1 × 131072 × 8192
129Add_21_ffnAdd1 × 131072 × 8192
130RMSNorm_22_1RMSNorm1 × 131072 × 8192
131Attention_22Grouped Query Attn1 × 131072 × 8192
132Add_22_attnAdd1 × 131072 × 8192
133RMSNorm_22_2RMSNorm1 × 131072 × 8192
134FFN_22SwiGLU1 × 131072 × 8192
135Add_22_ffnAdd1 × 131072 × 8192
136RMSNorm_23_1RMSNorm1 × 131072 × 8192
137Attention_23Grouped Query Attn1 × 131072 × 8192
138Add_23_attnAdd1 × 131072 × 8192
139RMSNorm_23_2RMSNorm1 × 131072 × 8192
140FFN_23SwiGLU1 × 131072 × 8192
141Add_23_ffnAdd1 × 131072 × 8192
142RMSNorm_24_1RMSNorm1 × 131072 × 8192
143Attention_24Grouped Query Attn1 × 131072 × 8192
144Add_24_attnAdd1 × 131072 × 8192
145RMSNorm_24_2RMSNorm1 × 131072 × 8192
146FFN_24SwiGLU1 × 131072 × 8192
147Add_24_ffnAdd1 × 131072 × 8192
148RMSNorm_25_1RMSNorm1 × 131072 × 8192
149Attention_25Grouped Query Attn1 × 131072 × 8192
150Add_25_attnAdd1 × 131072 × 8192
151RMSNorm_25_2RMSNorm1 × 131072 × 8192
152FFN_25SwiGLU1 × 131072 × 8192
153Add_25_ffnAdd1 × 131072 × 8192
154RMSNorm_26_1RMSNorm1 × 131072 × 8192
155Attention_26Grouped Query Attn1 × 131072 × 8192
156Add_26_attnAdd1 × 131072 × 8192
157RMSNorm_26_2RMSNorm1 × 131072 × 8192
158FFN_26SwiGLU1 × 131072 × 8192
159Add_26_ffnAdd1 × 131072 × 8192
160RMSNorm_27_1RMSNorm1 × 131072 × 8192
161Attention_27Grouped Query Attn1 × 131072 × 8192
162Add_27_attnAdd1 × 131072 × 8192
163RMSNorm_27_2RMSNorm1 × 131072 × 8192
164FFN_27SwiGLU1 × 131072 × 8192
165Add_27_ffnAdd1 × 131072 × 8192
166RMSNorm_28_1RMSNorm1 × 131072 × 8192
167Attention_28Grouped Query Attn1 × 131072 × 8192
168Add_28_attnAdd1 × 131072 × 8192
169RMSNorm_28_2RMSNorm1 × 131072 × 8192
170FFN_28SwiGLU1 × 131072 × 8192
171Add_28_ffnAdd1 × 131072 × 8192
172RMSNorm_29_1RMSNorm1 × 131072 × 8192
173Attention_29Grouped Query Attn1 × 131072 × 8192
174Add_29_attnAdd1 × 131072 × 8192
175RMSNorm_29_2RMSNorm1 × 131072 × 8192
176FFN_29SwiGLU1 × 131072 × 8192
177Add_29_ffnAdd1 × 131072 × 8192
178RMSNorm_30_1RMSNorm1 × 131072 × 8192
179Attention_30Grouped Query Attn1 × 131072 × 8192
180Add_30_attnAdd1 × 131072 × 8192
181RMSNorm_30_2RMSNorm1 × 131072 × 8192
182FFN_30SwiGLU1 × 131072 × 8192
183Add_30_ffnAdd1 × 131072 × 8192
184RMSNorm_31_1RMSNorm1 × 131072 × 8192
185Attention_31Grouped Query Attn1 × 131072 × 8192
186Add_31_attnAdd1 × 131072 × 8192
187RMSNorm_31_2RMSNorm1 × 131072 × 8192
188FFN_31SwiGLU1 × 131072 × 8192
189Add_31_ffnAdd1 × 131072 × 8192
190RMSNorm_32_1RMSNorm1 × 131072 × 8192
191Attention_32Grouped Query Attn1 × 131072 × 8192
192Add_32_attnAdd1 × 131072 × 8192
193RMSNorm_32_2RMSNorm1 × 131072 × 8192
194FFN_32SwiGLU1 × 131072 × 8192
195Add_32_ffnAdd1 × 131072 × 8192
196RMSNorm_33_1RMSNorm1 × 131072 × 8192
197Attention_33Grouped Query Attn1 × 131072 × 8192
198Add_33_attnAdd1 × 131072 × 8192
199RMSNorm_33_2RMSNorm1 × 131072 × 8192
200FFN_33SwiGLU1 × 131072 × 8192
201Add_33_ffnAdd1 × 131072 × 8192
202RMSNorm_34_1RMSNorm1 × 131072 × 8192
203Attention_34Grouped Query Attn1 × 131072 × 8192
204Add_34_attnAdd1 × 131072 × 8192
205RMSNorm_34_2RMSNorm1 × 131072 × 8192
206FFN_34SwiGLU1 × 131072 × 8192
207Add_34_ffnAdd1 × 131072 × 8192
208RMSNorm_35_1RMSNorm1 × 131072 × 8192
209Attention_35Grouped Query Attn1 × 131072 × 8192
210Add_35_attnAdd1 × 131072 × 8192
211RMSNorm_35_2RMSNorm1 × 131072 × 8192
212FFN_35SwiGLU1 × 131072 × 8192
213Add_35_ffnAdd1 × 131072 × 8192
214RMSNorm_36_1RMSNorm1 × 131072 × 8192
215Attention_36Grouped Query Attn1 × 131072 × 8192
216Add_36_attnAdd1 × 131072 × 8192
217RMSNorm_36_2RMSNorm1 × 131072 × 8192
218FFN_36SwiGLU1 × 131072 × 8192
219Add_36_ffnAdd1 × 131072 × 8192
220RMSNorm_37_1RMSNorm1 × 131072 × 8192
221Attention_37Grouped Query Attn1 × 131072 × 8192
222Add_37_attnAdd1 × 131072 × 8192
223RMSNorm_37_2RMSNorm1 × 131072 × 8192
224FFN_37SwiGLU1 × 131072 × 8192
225Add_37_ffnAdd1 × 131072 × 8192
226RMSNorm_38_1RMSNorm1 × 131072 × 8192
227Attention_38Grouped Query Attn1 × 131072 × 8192
228Add_38_attnAdd1 × 131072 × 8192
229RMSNorm_38_2RMSNorm1 × 131072 × 8192
230FFN_38SwiGLU1 × 131072 × 8192
231Add_38_ffnAdd1 × 131072 × 8192
232RMSNorm_39_1RMSNorm1 × 131072 × 8192
233Attention_39Grouped Query Attn1 × 131072 × 8192
234Add_39_attnAdd1 × 131072 × 8192
235RMSNorm_39_2RMSNorm1 × 131072 × 8192
236FFN_39SwiGLU1 × 131072 × 8192
237Add_39_ffnAdd1 × 131072 × 8192
238RMSNorm_40_1RMSNorm1 × 131072 × 8192
239Attention_40Grouped Query Attn1 × 131072 × 8192
240Add_40_attnAdd1 × 131072 × 8192
241RMSNorm_40_2RMSNorm1 × 131072 × 8192
242FFN_40SwiGLU1 × 131072 × 8192
243Add_40_ffnAdd1 × 131072 × 8192
244RMSNorm_41_1RMSNorm1 × 131072 × 8192
245Attention_41Grouped Query Attn1 × 131072 × 8192
246Add_41_attnAdd1 × 131072 × 8192
247RMSNorm_41_2RMSNorm1 × 131072 × 8192
248FFN_41SwiGLU1 × 131072 × 8192
249Add_41_ffnAdd1 × 131072 × 8192
250RMSNorm_42_1RMSNorm1 × 131072 × 8192
251Attention_42Grouped Query Attn1 × 131072 × 8192
252Add_42_attnAdd1 × 131072 × 8192
253RMSNorm_42_2RMSNorm1 × 131072 × 8192
254FFN_42SwiGLU1 × 131072 × 8192
255Add_42_ffnAdd1 × 131072 × 8192
256RMSNorm_43_1RMSNorm1 × 131072 × 8192
257Attention_43Grouped Query Attn1 × 131072 × 8192
258Add_43_attnAdd1 × 131072 × 8192
259RMSNorm_43_2RMSNorm1 × 131072 × 8192
260FFN_43SwiGLU1 × 131072 × 8192
261Add_43_ffnAdd1 × 131072 × 8192
262RMSNorm_44_1RMSNorm1 × 131072 × 8192
263Attention_44Grouped Query Attn1 × 131072 × 8192
264Add_44_attnAdd1 × 131072 × 8192
265RMSNorm_44_2RMSNorm1 × 131072 × 8192
266FFN_44SwiGLU1 × 131072 × 8192
267Add_44_ffnAdd1 × 131072 × 8192
268RMSNorm_45_1RMSNorm1 × 131072 × 8192
269Attention_45Grouped Query Attn1 × 131072 × 8192
270Add_45_attnAdd1 × 131072 × 8192
271RMSNorm_45_2RMSNorm1 × 131072 × 8192
272FFN_45SwiGLU1 × 131072 × 8192
273Add_45_ffnAdd1 × 131072 × 8192
274RMSNorm_46_1RMSNorm1 × 131072 × 8192
275Attention_46Grouped Query Attn1 × 131072 × 8192
276Add_46_attnAdd1 × 131072 × 8192
277RMSNorm_46_2RMSNorm1 × 131072 × 8192
278FFN_46SwiGLU1 × 131072 × 8192
279Add_46_ffnAdd1 × 131072 × 8192
280RMSNorm_47_1RMSNorm1 × 131072 × 8192
281Attention_47Grouped Query Attn1 × 131072 × 8192
282Add_47_attnAdd1 × 131072 × 8192
283RMSNorm_47_2RMSNorm1 × 131072 × 8192
284FFN_47SwiGLU1 × 131072 × 8192
285Add_47_ffnAdd1 × 131072 × 8192
286RMSNorm_48_1RMSNorm1 × 131072 × 8192
287Attention_48Grouped Query Attn1 × 131072 × 8192
288Add_48_attnAdd1 × 131072 × 8192
289RMSNorm_48_2RMSNorm1 × 131072 × 8192
290FFN_48SwiGLU1 × 131072 × 8192
291Add_48_ffnAdd1 × 131072 × 8192
292RMSNorm_49_1RMSNorm1 × 131072 × 8192
293Attention_49Grouped Query Attn1 × 131072 × 8192
294Add_49_attnAdd1 × 131072 × 8192
295RMSNorm_49_2RMSNorm1 × 131072 × 8192
296FFN_49SwiGLU1 × 131072 × 8192
297Add_49_ffnAdd1 × 131072 × 8192
298RMSNorm_50_1RMSNorm1 × 131072 × 8192
299Attention_50Grouped Query Attn1 × 131072 × 8192
300Add_50_attnAdd1 × 131072 × 8192
301RMSNorm_50_2RMSNorm1 × 131072 × 8192
302FFN_50SwiGLU1 × 131072 × 8192
303Add_50_ffnAdd1 × 131072 × 8192
304RMSNorm_51_1RMSNorm1 × 131072 × 8192
305Attention_51Grouped Query Attn1 × 131072 × 8192
306Add_51_attnAdd1 × 131072 × 8192
307RMSNorm_51_2RMSNorm1 × 131072 × 8192
308FFN_51SwiGLU1 × 131072 × 8192
309Add_51_ffnAdd1 × 131072 × 8192
310RMSNorm_52_1RMSNorm1 × 131072 × 8192
311Attention_52Grouped Query Attn1 × 131072 × 8192
312Add_52_attnAdd1 × 131072 × 8192
313RMSNorm_52_2RMSNorm1 × 131072 × 8192
314FFN_52SwiGLU1 × 131072 × 8192
315Add_52_ffnAdd1 × 131072 × 8192
316RMSNorm_53_1RMSNorm1 × 131072 × 8192
317Attention_53Grouped Query Attn1 × 131072 × 8192
318Add_53_attnAdd1 × 131072 × 8192
319RMSNorm_53_2RMSNorm1 × 131072 × 8192
320FFN_53SwiGLU1 × 131072 × 8192
321Add_53_ffnAdd1 × 131072 × 8192
322RMSNorm_54_1RMSNorm1 × 131072 × 8192
323Attention_54Grouped Query Attn1 × 131072 × 8192
324Add_54_attnAdd1 × 131072 × 8192
325RMSNorm_54_2RMSNorm1 × 131072 × 8192
326FFN_54SwiGLU1 × 131072 × 8192
327Add_54_ffnAdd1 × 131072 × 8192
328RMSNorm_55_1RMSNorm1 × 131072 × 8192
329Attention_55Grouped Query Attn1 × 131072 × 8192
330Add_55_attnAdd1 × 131072 × 8192
331RMSNorm_55_2RMSNorm1 × 131072 × 8192
332FFN_55SwiGLU1 × 131072 × 8192
333Add_55_ffnAdd1 × 131072 × 8192
334RMSNorm_56_1RMSNorm1 × 131072 × 8192
335Attention_56Grouped Query Attn1 × 131072 × 8192
336Add_56_attnAdd1 × 131072 × 8192
337RMSNorm_56_2RMSNorm1 × 131072 × 8192
338FFN_56SwiGLU1 × 131072 × 8192
339Add_56_ffnAdd1 × 131072 × 8192
340RMSNorm_57_1RMSNorm1 × 131072 × 8192
341Attention_57Grouped Query Attn1 × 131072 × 8192
342Add_57_attnAdd1 × 131072 × 8192
343RMSNorm_57_2RMSNorm1 × 131072 × 8192
344FFN_57SwiGLU1 × 131072 × 8192
345Add_57_ffnAdd1 × 131072 × 8192
346RMSNorm_58_1RMSNorm1 × 131072 × 8192
347Attention_58Grouped Query Attn1 × 131072 × 8192
348Add_58_attnAdd1 × 131072 × 8192
349RMSNorm_58_2RMSNorm1 × 131072 × 8192
350FFN_58SwiGLU1 × 131072 × 8192
351Add_58_ffnAdd1 × 131072 × 8192
352RMSNorm_59_1RMSNorm1 × 131072 × 8192
353Attention_59Grouped Query Attn1 × 131072 × 8192
354Add_59_attnAdd1 × 131072 × 8192
355RMSNorm_59_2RMSNorm1 × 131072 × 8192
356FFN_59SwiGLU1 × 131072 × 8192
357Add_59_ffnAdd1 × 131072 × 8192
358RMSNorm_60_1RMSNorm1 × 131072 × 8192
359Attention_60Grouped Query Attn1 × 131072 × 8192
360Add_60_attnAdd1 × 131072 × 8192
361RMSNorm_60_2RMSNorm1 × 131072 × 8192
362FFN_60SwiGLU1 × 131072 × 8192
363Add_60_ffnAdd1 × 131072 × 8192
364RMSNorm_61_1RMSNorm1 × 131072 × 8192
365Attention_61Grouped Query Attn1 × 131072 × 8192
366Add_61_attnAdd1 × 131072 × 8192
367RMSNorm_61_2RMSNorm1 × 131072 × 8192
368FFN_61SwiGLU1 × 131072 × 8192
369Add_61_ffnAdd1 × 131072 × 8192
370RMSNorm_62_1RMSNorm1 × 131072 × 8192
371Attention_62Grouped Query Attn1 × 131072 × 8192
372Add_62_attnAdd1 × 131072 × 8192
373RMSNorm_62_2RMSNorm1 × 131072 × 8192
374FFN_62SwiGLU1 × 131072 × 8192
375Add_62_ffnAdd1 × 131072 × 8192
376RMSNorm_63_1RMSNorm1 × 131072 × 8192
377Attention_63Grouped Query Attn1 × 131072 × 8192
378Add_63_attnAdd1 × 131072 × 8192
379RMSNorm_63_2RMSNorm1 × 131072 × 8192
380FFN_63SwiGLU1 × 131072 × 8192
381Add_63_ffnAdd1 × 131072 × 8192
382RMSNorm_64_1RMSNorm1 × 131072 × 8192
383Attention_64Grouped Query Attn1 × 131072 × 8192
384Add_64_attnAdd1 × 131072 × 8192
385RMSNorm_64_2RMSNorm1 × 131072 × 8192
386FFN_64SwiGLU1 × 131072 × 8192
387Add_64_ffnAdd1 × 131072 × 8192
388RMSNorm_65_1RMSNorm1 × 131072 × 8192
389Attention_65Grouped Query Attn1 × 131072 × 8192
390Add_65_attnAdd1 × 131072 × 8192
391RMSNorm_65_2RMSNorm1 × 131072 × 8192
392FFN_65SwiGLU1 × 131072 × 8192
393Add_65_ffnAdd1 × 131072 × 8192
394RMSNorm_66_1RMSNorm1 × 131072 × 8192
395Attention_66Grouped Query Attn1 × 131072 × 8192
396Add_66_attnAdd1 × 131072 × 8192
397RMSNorm_66_2RMSNorm1 × 131072 × 8192
398FFN_66SwiGLU1 × 131072 × 8192
399Add_66_ffnAdd1 × 131072 × 8192
400RMSNorm_67_1RMSNorm1 × 131072 × 8192
401Attention_67Grouped Query Attn1 × 131072 × 8192
402Add_67_attnAdd1 × 131072 × 8192
403RMSNorm_67_2RMSNorm1 × 131072 × 8192
404FFN_67SwiGLU1 × 131072 × 8192
405Add_67_ffnAdd1 × 131072 × 8192
406RMSNorm_68_1RMSNorm1 × 131072 × 8192
407Attention_68Grouped Query Attn1 × 131072 × 8192
408Add_68_attnAdd1 × 131072 × 8192
409RMSNorm_68_2RMSNorm1 × 131072 × 8192
410FFN_68SwiGLU1 × 131072 × 8192
411Add_68_ffnAdd1 × 131072 × 8192
412RMSNorm_69_1RMSNorm1 × 131072 × 8192
413Attention_69Grouped Query Attn1 × 131072 × 8192
414Add_69_attnAdd1 × 131072 × 8192
415RMSNorm_69_2RMSNorm1 × 131072 × 8192
416FFN_69SwiGLU1 × 131072 × 8192
417Add_69_ffnAdd1 × 131072 × 8192
418RMSNorm_70_1RMSNorm1 × 131072 × 8192
419Attention_70Grouped Query Attn1 × 131072 × 8192
420Add_70_attnAdd1 × 131072 × 8192
421RMSNorm_70_2RMSNorm1 × 131072 × 8192
422FFN_70SwiGLU1 × 131072 × 8192
423Add_70_ffnAdd1 × 131072 × 8192
424RMSNorm_71_1RMSNorm1 × 131072 × 8192
425Attention_71Grouped Query Attn1 × 131072 × 8192
426Add_71_attnAdd1 × 131072 × 8192
427RMSNorm_71_2RMSNorm1 × 131072 × 8192
428FFN_71SwiGLU1 × 131072 × 8192
429Add_71_ffnAdd1 × 131072 × 8192
430RMSNorm_72_1RMSNorm1 × 131072 × 8192
431Attention_72Grouped Query Attn1 × 131072 × 8192
432Add_72_attnAdd1 × 131072 × 8192
433RMSNorm_72_2RMSNorm1 × 131072 × 8192
434FFN_72SwiGLU1 × 131072 × 8192
435Add_72_ffnAdd1 × 131072 × 8192
436RMSNorm_73_1RMSNorm1 × 131072 × 8192
437Attention_73Grouped Query Attn1 × 131072 × 8192
438Add_73_attnAdd1 × 131072 × 8192
439RMSNorm_73_2RMSNorm1 × 131072 × 8192
440FFN_73SwiGLU1 × 131072 × 8192
441Add_73_ffnAdd1 × 131072 × 8192
442RMSNorm_74_1RMSNorm1 × 131072 × 8192
443Attention_74Grouped Query Attn1 × 131072 × 8192
444Add_74_attnAdd1 × 131072 × 8192
445RMSNorm_74_2RMSNorm1 × 131072 × 8192
446FFN_74SwiGLU1 × 131072 × 8192
447Add_74_ffnAdd1 × 131072 × 8192
448RMSNorm_75_1RMSNorm1 × 131072 × 8192
449Attention_75Grouped Query Attn1 × 131072 × 8192
450Add_75_attnAdd1 × 131072 × 8192
451RMSNorm_75_2RMSNorm1 × 131072 × 8192
452FFN_75SwiGLU1 × 131072 × 8192
453Add_75_ffnAdd1 × 131072 × 8192
454RMSNorm_76_1RMSNorm1 × 131072 × 8192
455Attention_76Grouped Query Attn1 × 131072 × 8192
456Add_76_attnAdd1 × 131072 × 8192
457RMSNorm_76_2RMSNorm1 × 131072 × 8192
458FFN_76SwiGLU1 × 131072 × 8192
459Add_76_ffnAdd1 × 131072 × 8192
460RMSNorm_77_1RMSNorm1 × 131072 × 8192
461Attention_77Grouped Query Attn1 × 131072 × 8192
462Add_77_attnAdd1 × 131072 × 8192
463RMSNorm_77_2RMSNorm1 × 131072 × 8192
464FFN_77SwiGLU1 × 131072 × 8192
465Add_77_ffnAdd1 × 131072 × 8192
466RMSNorm_78_1RMSNorm1 × 131072 × 8192
467Attention_78Grouped Query Attn1 × 131072 × 8192
468Add_78_attnAdd1 × 131072 × 8192
469RMSNorm_78_2RMSNorm1 × 131072 × 8192
470FFN_78SwiGLU1 × 131072 × 8192
471Add_78_ffnAdd1 × 131072 × 8192
472RMSNorm_79_1RMSNorm1 × 131072 × 8192
473Attention_79Grouped Query Attn1 × 131072 × 8192
474Add_79_attnAdd1 × 131072 × 8192
475RMSNorm_79_2RMSNorm1 × 131072 × 8192
476FFN_79SwiGLU1 × 131072 × 8192
477Add_79_ffnAdd1 × 131072 × 8192
478RMSNorm_80_1RMSNorm1 × 131072 × 8192
479Attention_80Grouped Query Attn1 × 131072 × 8192
480Add_80_attnAdd1 × 131072 × 8192
481RMSNorm_80_2RMSNorm1 × 131072 × 8192
482FFN_80SwiGLU1 × 131072 × 8192
483Add_80_ffnAdd1 × 131072 × 8192
484Final_RMSNormRMSNorm1 × 131072 × 8192
485LM_HeadLinear1 × 131072 × 128256
486OutputOutput1 × 131072 × 128256

What the verifier says

infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 3072 (0.38× embedDim). Expected: ~21760. Fix: Set intermediateSize to 21760 for embedDim=8192.
swiglu-dim-convention
warnLinear "LM_Head" is 8192 × 128256 = 1051M parameters (~3.9 GB float32). A single dense layer this large usually means a feature map was flattened without pooling first; embedding / vocab-projection heads are the expected exception. Fix: Add a Global Average Pool or more downsampling before the Linear, or factorize it (low-rank / bottleneck projection).
huge-linear-params
info80 attention layers at embedDim 8192 cache full per-head K/V: about 2560 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 80 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 80 attention layers this design caches 2560 KB per token, so a single 8,192-token sequence needs ~21.5 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/Llama-3_3-Nemotron-Super-49B-v1 --plan --share

Other nemotron-nas checkpoints

Llama-3_3-Nemotron-Super-49B-v1_5-FP8
29.62B derived · -40.6% against the checkpoint