N Neurarch Architectures Models Checks Data Docs Open the app

Models / gemma4

OTel-2.0-LLM-31B-IT

Reconstructed from its own config.json with no weights read. 6.8M 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
30.56B
30,564,872,304 parameters
In the published checkpoint
31.27B
31,273,088,876 scalars · safetensors.total, read 2026-09-06
Delta
-2.26%

multi-tower The config declares 2 sub-models (text_config, vision_config). The published checkpoint carries all of them; the graph below carries the towers the reader reconstructs. 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
475
Will it forward-pass
Yes
Priced on
A10G (24GB)
Est. one run
$1548238.17
CardMemory
T4 (16GB)weights + activationsdoes not fit
A100 (40GB)weights + activationsdoes not fit
H100 (80GB)weights + activationsdoes not fit

Structure

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

LayerTypeOutput shape
1InputInput1 × 262144
2EmbeddingEmbedding1 × 262144 × 5376
3RoPERoPE1 × 262144 × 5376
4Vision inputInput3 × 224 × 224
5PatchEmbedPatch Embed196 × 1152
6Patch_Position_EmbeddingLearned Pos Embed196 × 1152
7Vision_LN_1LayerNorm196 × 1152
8Vision_Attn_1Multi-Head Attention196 × 1152
9Vision_Add_1Add196 × 1152
10Vision_FFN_1Feed Forward196 × 1152
11Vision_LN_2LayerNorm196 × 1152
12Vision_Attn_2Multi-Head Attention196 × 1152
13Vision_Add_2Add196 × 1152
14Vision_FFN_2Feed Forward196 × 1152
15Vision_LN_3LayerNorm196 × 1152
16Vision_Attn_3Multi-Head Attention196 × 1152
17Vision_Add_3Add196 × 1152
18Vision_FFN_3Feed Forward196 × 1152
19Vision_LN_4LayerNorm196 × 1152
20Vision_Attn_4Multi-Head Attention196 × 1152
21Vision_Add_4Add196 × 1152
22Vision_FFN_4Feed Forward196 × 1152
23Vision_LN_5LayerNorm196 × 1152
24Vision_Attn_5Multi-Head Attention196 × 1152
25Vision_Add_5Add196 × 1152
26Vision_FFN_5Feed Forward196 × 1152
27Vision_LN_6LayerNorm196 × 1152
28Vision_Attn_6Multi-Head Attention196 × 1152
29Vision_Add_6Add196 × 1152
30Vision_FFN_6Feed Forward196 × 1152
31Vision_LN_7LayerNorm196 × 1152
32Vision_Attn_7Multi-Head Attention196 × 1152
33Vision_Add_7Add196 × 1152
34Vision_FFN_7Feed Forward196 × 1152
35Vision_LN_8LayerNorm196 × 1152
36Vision_Attn_8Multi-Head Attention196 × 1152
37Vision_Add_8Add196 × 1152
38Vision_FFN_8Feed Forward196 × 1152
39Vision_LN_9LayerNorm196 × 1152
40Vision_Attn_9Multi-Head Attention196 × 1152
41Vision_Add_9Add196 × 1152
42Vision_FFN_9Feed Forward196 × 1152
43Vision_LN_10LayerNorm196 × 1152
44Vision_Attn_10Multi-Head Attention196 × 1152
45Vision_Add_10Add196 × 1152
46Vision_FFN_10Feed Forward196 × 1152
47Vision_LN_11LayerNorm196 × 1152
48Vision_Attn_11Multi-Head Attention196 × 1152
49Vision_Add_11Add196 × 1152
50Vision_FFN_11Feed Forward196 × 1152
51Vision_LN_12LayerNorm196 × 1152
52Vision_Attn_12Multi-Head Attention196 × 1152
53Vision_Add_12Add196 × 1152
54Vision_FFN_12Feed Forward196 × 1152
55Vision_LN_13LayerNorm196 × 1152
56Vision_Attn_13Multi-Head Attention196 × 1152
57Vision_Add_13Add196 × 1152
58Vision_FFN_13Feed Forward196 × 1152
59Vision_LN_14LayerNorm196 × 1152
60Vision_Attn_14Multi-Head Attention196 × 1152
61Vision_Add_14Add196 × 1152
62Vision_FFN_14Feed Forward196 × 1152
63Vision_LN_15LayerNorm196 × 1152
64Vision_Attn_15Multi-Head Attention196 × 1152
65Vision_Add_15Add196 × 1152
66Vision_FFN_15Feed Forward196 × 1152
67Vision_LN_16LayerNorm196 × 1152
68Vision_Attn_16Multi-Head Attention196 × 1152
69Vision_Add_16Add196 × 1152
70Vision_FFN_16Feed Forward196 × 1152
71Vision_LN_17LayerNorm196 × 1152
72Vision_Attn_17Multi-Head Attention196 × 1152
73Vision_Add_17Add196 × 1152
74Vision_FFN_17Feed Forward196 × 1152
75Vision_LN_18LayerNorm196 × 1152
76Vision_Attn_18Multi-Head Attention196 × 1152
77Vision_Add_18Add196 × 1152
78Vision_FFN_18Feed Forward196 × 1152
79Vision_LN_19LayerNorm196 × 1152
80Vision_Attn_19Multi-Head Attention196 × 1152
81Vision_Add_19Add196 × 1152
82Vision_FFN_19Feed Forward196 × 1152
83Vision_LN_20LayerNorm196 × 1152
84Vision_Attn_20Multi-Head Attention196 × 1152
85Vision_Add_20Add196 × 1152
86Vision_FFN_20Feed Forward196 × 1152
87Vision_LN_21LayerNorm196 × 1152
88Vision_Attn_21Multi-Head Attention196 × 1152
89Vision_Add_21Add196 × 1152
90Vision_FFN_21Feed Forward196 × 1152
91Vision_LN_22LayerNorm196 × 1152
92Vision_Attn_22Multi-Head Attention196 × 1152
93Vision_Add_22Add196 × 1152
94Vision_FFN_22Feed Forward196 × 1152
95Vision_LN_23LayerNorm196 × 1152
96Vision_Attn_23Multi-Head Attention196 × 1152
97Vision_Add_23Add196 × 1152
98Vision_FFN_23Feed Forward196 × 1152
99Vision_LN_24LayerNorm196 × 1152
100Vision_Attn_24Multi-Head Attention196 × 1152
101Vision_Add_24Add196 × 1152
102Vision_FFN_24Feed Forward196 × 1152
103Vision_LN_25LayerNorm196 × 1152
104Vision_Attn_25Multi-Head Attention196 × 1152
105Vision_Add_25Add196 × 1152
106Vision_FFN_25Feed Forward196 × 1152
107Vision_LN_26LayerNorm196 × 1152
108Vision_Attn_26Multi-Head Attention196 × 1152
109Vision_Add_26Add196 × 1152
110Vision_FFN_26Feed Forward196 × 1152
111Vision_LN_27LayerNorm196 × 1152
112Vision_Attn_27Multi-Head Attention196 × 1152
113Vision_Add_27Add196 × 1152
114Vision_FFN_27Feed Forward196 × 1152
115Vision projectorProjection196 × 5376
116Vision tokensReshape1 × 196 × 5376
117Multimodal fusion (concat tokens)Concatenate1 × 262340 × 5376
118RMSNorm_1_1RMSNorm1 × 262340 × 5376
119Attention_1Grouped Query Attn1 × 262340 × 5376
120Add_1_attnAdd1 × 262340 × 5376
121RMSNorm_1_2RMSNorm1 × 262340 × 5376
122FFN_1SwiGLU1 × 262340 × 5376
123Add_1_ffnAdd1 × 262340 × 5376
124RMSNorm_2_1RMSNorm1 × 262340 × 5376
125Attention_2Grouped Query Attn1 × 262340 × 5376
126Add_2_attnAdd1 × 262340 × 5376
127RMSNorm_2_2RMSNorm1 × 262340 × 5376
128FFN_2SwiGLU1 × 262340 × 5376
129Add_2_ffnAdd1 × 262340 × 5376
130RMSNorm_3_1RMSNorm1 × 262340 × 5376
131Attention_3Grouped Query Attn1 × 262340 × 5376
132Add_3_attnAdd1 × 262340 × 5376
133RMSNorm_3_2RMSNorm1 × 262340 × 5376
134FFN_3SwiGLU1 × 262340 × 5376
135Add_3_ffnAdd1 × 262340 × 5376
136RMSNorm_4_1RMSNorm1 × 262340 × 5376
137Attention_4Grouped Query Attn1 × 262340 × 5376
138Add_4_attnAdd1 × 262340 × 5376
139RMSNorm_4_2RMSNorm1 × 262340 × 5376
140FFN_4SwiGLU1 × 262340 × 5376
141Add_4_ffnAdd1 × 262340 × 5376
142RMSNorm_5_1RMSNorm1 × 262340 × 5376
143Attention_5Grouped Query Attn1 × 262340 × 5376
144Add_5_attnAdd1 × 262340 × 5376
145RMSNorm_5_2RMSNorm1 × 262340 × 5376
146FFN_5SwiGLU1 × 262340 × 5376
147Add_5_ffnAdd1 × 262340 × 5376
148RMSNorm_6_1RMSNorm1 × 262340 × 5376
149Attention_6Grouped Query Attn1 × 262340 × 5376
150Add_6_attnAdd1 × 262340 × 5376
151RMSNorm_6_2RMSNorm1 × 262340 × 5376
152FFN_6SwiGLU1 × 262340 × 5376
153Add_6_ffnAdd1 × 262340 × 5376
154RMSNorm_7_1RMSNorm1 × 262340 × 5376
155Attention_7Grouped Query Attn1 × 262340 × 5376
156Add_7_attnAdd1 × 262340 × 5376
157RMSNorm_7_2RMSNorm1 × 262340 × 5376
158FFN_7SwiGLU1 × 262340 × 5376
159Add_7_ffnAdd1 × 262340 × 5376
160RMSNorm_8_1RMSNorm1 × 262340 × 5376
161Attention_8Grouped Query Attn1 × 262340 × 5376
162Add_8_attnAdd1 × 262340 × 5376
163RMSNorm_8_2RMSNorm1 × 262340 × 5376
164FFN_8SwiGLU1 × 262340 × 5376
165Add_8_ffnAdd1 × 262340 × 5376
166RMSNorm_9_1RMSNorm1 × 262340 × 5376
167Attention_9Grouped Query Attn1 × 262340 × 5376
168Add_9_attnAdd1 × 262340 × 5376
169RMSNorm_9_2RMSNorm1 × 262340 × 5376
170FFN_9SwiGLU1 × 262340 × 5376
171Add_9_ffnAdd1 × 262340 × 5376
172RMSNorm_10_1RMSNorm1 × 262340 × 5376
173Attention_10Grouped Query Attn1 × 262340 × 5376
174Add_10_attnAdd1 × 262340 × 5376
175RMSNorm_10_2RMSNorm1 × 262340 × 5376
176FFN_10SwiGLU1 × 262340 × 5376
177Add_10_ffnAdd1 × 262340 × 5376
178RMSNorm_11_1RMSNorm1 × 262340 × 5376
179Attention_11Grouped Query Attn1 × 262340 × 5376
180Add_11_attnAdd1 × 262340 × 5376
181RMSNorm_11_2RMSNorm1 × 262340 × 5376
182FFN_11SwiGLU1 × 262340 × 5376
183Add_11_ffnAdd1 × 262340 × 5376
184RMSNorm_12_1RMSNorm1 × 262340 × 5376
185Attention_12Grouped Query Attn1 × 262340 × 5376
186Add_12_attnAdd1 × 262340 × 5376
187RMSNorm_12_2RMSNorm1 × 262340 × 5376
188FFN_12SwiGLU1 × 262340 × 5376
189Add_12_ffnAdd1 × 262340 × 5376
190RMSNorm_13_1RMSNorm1 × 262340 × 5376
191Attention_13Grouped Query Attn1 × 262340 × 5376
192Add_13_attnAdd1 × 262340 × 5376
193RMSNorm_13_2RMSNorm1 × 262340 × 5376
194FFN_13SwiGLU1 × 262340 × 5376
195Add_13_ffnAdd1 × 262340 × 5376
196RMSNorm_14_1RMSNorm1 × 262340 × 5376
197Attention_14Grouped Query Attn1 × 262340 × 5376
198Add_14_attnAdd1 × 262340 × 5376
199RMSNorm_14_2RMSNorm1 × 262340 × 5376
200FFN_14SwiGLU1 × 262340 × 5376
201Add_14_ffnAdd1 × 262340 × 5376
202RMSNorm_15_1RMSNorm1 × 262340 × 5376
203Attention_15Grouped Query Attn1 × 262340 × 5376
204Add_15_attnAdd1 × 262340 × 5376
205RMSNorm_15_2RMSNorm1 × 262340 × 5376
206FFN_15SwiGLU1 × 262340 × 5376
207Add_15_ffnAdd1 × 262340 × 5376
208RMSNorm_16_1RMSNorm1 × 262340 × 5376
209Attention_16Grouped Query Attn1 × 262340 × 5376
210Add_16_attnAdd1 × 262340 × 5376
211RMSNorm_16_2RMSNorm1 × 262340 × 5376
212FFN_16SwiGLU1 × 262340 × 5376
213Add_16_ffnAdd1 × 262340 × 5376
214RMSNorm_17_1RMSNorm1 × 262340 × 5376
215Attention_17Grouped Query Attn1 × 262340 × 5376
216Add_17_attnAdd1 × 262340 × 5376
217RMSNorm_17_2RMSNorm1 × 262340 × 5376
218FFN_17SwiGLU1 × 262340 × 5376
219Add_17_ffnAdd1 × 262340 × 5376
220RMSNorm_18_1RMSNorm1 × 262340 × 5376
221Attention_18Grouped Query Attn1 × 262340 × 5376
222Add_18_attnAdd1 × 262340 × 5376
223RMSNorm_18_2RMSNorm1 × 262340 × 5376
224FFN_18SwiGLU1 × 262340 × 5376
225Add_18_ffnAdd1 × 262340 × 5376
226RMSNorm_19_1RMSNorm1 × 262340 × 5376
227Attention_19Grouped Query Attn1 × 262340 × 5376
228Add_19_attnAdd1 × 262340 × 5376
229RMSNorm_19_2RMSNorm1 × 262340 × 5376
230FFN_19SwiGLU1 × 262340 × 5376
231Add_19_ffnAdd1 × 262340 × 5376
232RMSNorm_20_1RMSNorm1 × 262340 × 5376
233Attention_20Grouped Query Attn1 × 262340 × 5376
234Add_20_attnAdd1 × 262340 × 5376
235RMSNorm_20_2RMSNorm1 × 262340 × 5376
236FFN_20SwiGLU1 × 262340 × 5376
237Add_20_ffnAdd1 × 262340 × 5376
238RMSNorm_21_1RMSNorm1 × 262340 × 5376
239Attention_21Grouped Query Attn1 × 262340 × 5376
240Add_21_attnAdd1 × 262340 × 5376
241RMSNorm_21_2RMSNorm1 × 262340 × 5376
242FFN_21SwiGLU1 × 262340 × 5376
243Add_21_ffnAdd1 × 262340 × 5376
244RMSNorm_22_1RMSNorm1 × 262340 × 5376
245Attention_22Grouped Query Attn1 × 262340 × 5376
246Add_22_attnAdd1 × 262340 × 5376
247RMSNorm_22_2RMSNorm1 × 262340 × 5376
248FFN_22SwiGLU1 × 262340 × 5376
249Add_22_ffnAdd1 × 262340 × 5376
250RMSNorm_23_1RMSNorm1 × 262340 × 5376
251Attention_23Grouped Query Attn1 × 262340 × 5376
252Add_23_attnAdd1 × 262340 × 5376
253RMSNorm_23_2RMSNorm1 × 262340 × 5376
254FFN_23SwiGLU1 × 262340 × 5376
255Add_23_ffnAdd1 × 262340 × 5376
256RMSNorm_24_1RMSNorm1 × 262340 × 5376
257Attention_24Grouped Query Attn1 × 262340 × 5376
258Add_24_attnAdd1 × 262340 × 5376
259RMSNorm_24_2RMSNorm1 × 262340 × 5376
260FFN_24SwiGLU1 × 262340 × 5376
261Add_24_ffnAdd1 × 262340 × 5376
262RMSNorm_25_1RMSNorm1 × 262340 × 5376
263Attention_25Grouped Query Attn1 × 262340 × 5376
264Add_25_attnAdd1 × 262340 × 5376
265RMSNorm_25_2RMSNorm1 × 262340 × 5376
266FFN_25SwiGLU1 × 262340 × 5376
267Add_25_ffnAdd1 × 262340 × 5376
268RMSNorm_26_1RMSNorm1 × 262340 × 5376
269Attention_26Grouped Query Attn1 × 262340 × 5376
270Add_26_attnAdd1 × 262340 × 5376
271RMSNorm_26_2RMSNorm1 × 262340 × 5376
272FFN_26SwiGLU1 × 262340 × 5376
273Add_26_ffnAdd1 × 262340 × 5376
274RMSNorm_27_1RMSNorm1 × 262340 × 5376
275Attention_27Grouped Query Attn1 × 262340 × 5376
276Add_27_attnAdd1 × 262340 × 5376
277RMSNorm_27_2RMSNorm1 × 262340 × 5376
278FFN_27SwiGLU1 × 262340 × 5376
279Add_27_ffnAdd1 × 262340 × 5376
280RMSNorm_28_1RMSNorm1 × 262340 × 5376
281Attention_28Grouped Query Attn1 × 262340 × 5376
282Add_28_attnAdd1 × 262340 × 5376
283RMSNorm_28_2RMSNorm1 × 262340 × 5376
284FFN_28SwiGLU1 × 262340 × 5376
285Add_28_ffnAdd1 × 262340 × 5376
286RMSNorm_29_1RMSNorm1 × 262340 × 5376
287Attention_29Grouped Query Attn1 × 262340 × 5376
288Add_29_attnAdd1 × 262340 × 5376
289RMSNorm_29_2RMSNorm1 × 262340 × 5376
290FFN_29SwiGLU1 × 262340 × 5376
291Add_29_ffnAdd1 × 262340 × 5376
292RMSNorm_30_1RMSNorm1 × 262340 × 5376
293Attention_30Grouped Query Attn1 × 262340 × 5376
294Add_30_attnAdd1 × 262340 × 5376
295RMSNorm_30_2RMSNorm1 × 262340 × 5376
296FFN_30SwiGLU1 × 262340 × 5376
297Add_30_ffnAdd1 × 262340 × 5376
298RMSNorm_31_1RMSNorm1 × 262340 × 5376
299Attention_31Grouped Query Attn1 × 262340 × 5376
300Add_31_attnAdd1 × 262340 × 5376
301RMSNorm_31_2RMSNorm1 × 262340 × 5376
302FFN_31SwiGLU1 × 262340 × 5376
303Add_31_ffnAdd1 × 262340 × 5376
304RMSNorm_32_1RMSNorm1 × 262340 × 5376
305Attention_32Grouped Query Attn1 × 262340 × 5376
306Add_32_attnAdd1 × 262340 × 5376
307RMSNorm_32_2RMSNorm1 × 262340 × 5376
308FFN_32SwiGLU1 × 262340 × 5376
309Add_32_ffnAdd1 × 262340 × 5376
310RMSNorm_33_1RMSNorm1 × 262340 × 5376
311Attention_33Grouped Query Attn1 × 262340 × 5376
312Add_33_attnAdd1 × 262340 × 5376
313RMSNorm_33_2RMSNorm1 × 262340 × 5376
314FFN_33SwiGLU1 × 262340 × 5376
315Add_33_ffnAdd1 × 262340 × 5376
316RMSNorm_34_1RMSNorm1 × 262340 × 5376
317Attention_34Grouped Query Attn1 × 262340 × 5376
318Add_34_attnAdd1 × 262340 × 5376
319RMSNorm_34_2RMSNorm1 × 262340 × 5376
320FFN_34SwiGLU1 × 262340 × 5376
321Add_34_ffnAdd1 × 262340 × 5376
322RMSNorm_35_1RMSNorm1 × 262340 × 5376
323Attention_35Grouped Query Attn1 × 262340 × 5376
324Add_35_attnAdd1 × 262340 × 5376
325RMSNorm_35_2RMSNorm1 × 262340 × 5376
326FFN_35SwiGLU1 × 262340 × 5376
327Add_35_ffnAdd1 × 262340 × 5376
328RMSNorm_36_1RMSNorm1 × 262340 × 5376
329Attention_36Grouped Query Attn1 × 262340 × 5376
330Add_36_attnAdd1 × 262340 × 5376
331RMSNorm_36_2RMSNorm1 × 262340 × 5376
332FFN_36SwiGLU1 × 262340 × 5376
333Add_36_ffnAdd1 × 262340 × 5376
334RMSNorm_37_1RMSNorm1 × 262340 × 5376
335Attention_37Grouped Query Attn1 × 262340 × 5376
336Add_37_attnAdd1 × 262340 × 5376
337RMSNorm_37_2RMSNorm1 × 262340 × 5376
338FFN_37SwiGLU1 × 262340 × 5376
339Add_37_ffnAdd1 × 262340 × 5376
340RMSNorm_38_1RMSNorm1 × 262340 × 5376
341Attention_38Grouped Query Attn1 × 262340 × 5376
342Add_38_attnAdd1 × 262340 × 5376
343RMSNorm_38_2RMSNorm1 × 262340 × 5376
344FFN_38SwiGLU1 × 262340 × 5376
345Add_38_ffnAdd1 × 262340 × 5376
346RMSNorm_39_1RMSNorm1 × 262340 × 5376
347Attention_39Grouped Query Attn1 × 262340 × 5376
348Add_39_attnAdd1 × 262340 × 5376
349RMSNorm_39_2RMSNorm1 × 262340 × 5376
350FFN_39SwiGLU1 × 262340 × 5376
351Add_39_ffnAdd1 × 262340 × 5376
352RMSNorm_40_1RMSNorm1 × 262340 × 5376
353Attention_40Grouped Query Attn1 × 262340 × 5376
354Add_40_attnAdd1 × 262340 × 5376
355RMSNorm_40_2RMSNorm1 × 262340 × 5376
356FFN_40SwiGLU1 × 262340 × 5376
357Add_40_ffnAdd1 × 262340 × 5376
358RMSNorm_41_1RMSNorm1 × 262340 × 5376
359Attention_41Grouped Query Attn1 × 262340 × 5376
360Add_41_attnAdd1 × 262340 × 5376
361RMSNorm_41_2RMSNorm1 × 262340 × 5376
362FFN_41SwiGLU1 × 262340 × 5376
363Add_41_ffnAdd1 × 262340 × 5376
364RMSNorm_42_1RMSNorm1 × 262340 × 5376
365Attention_42Grouped Query Attn1 × 262340 × 5376
366Add_42_attnAdd1 × 262340 × 5376
367RMSNorm_42_2RMSNorm1 × 262340 × 5376
368FFN_42SwiGLU1 × 262340 × 5376
369Add_42_ffnAdd1 × 262340 × 5376
370RMSNorm_43_1RMSNorm1 × 262340 × 5376
371Attention_43Grouped Query Attn1 × 262340 × 5376
372Add_43_attnAdd1 × 262340 × 5376
373RMSNorm_43_2RMSNorm1 × 262340 × 5376
374FFN_43SwiGLU1 × 262340 × 5376
375Add_43_ffnAdd1 × 262340 × 5376
376RMSNorm_44_1RMSNorm1 × 262340 × 5376
377Attention_44Grouped Query Attn1 × 262340 × 5376
378Add_44_attnAdd1 × 262340 × 5376
379RMSNorm_44_2RMSNorm1 × 262340 × 5376
380FFN_44SwiGLU1 × 262340 × 5376
381Add_44_ffnAdd1 × 262340 × 5376
382RMSNorm_45_1RMSNorm1 × 262340 × 5376
383Attention_45Grouped Query Attn1 × 262340 × 5376
384Add_45_attnAdd1 × 262340 × 5376
385RMSNorm_45_2RMSNorm1 × 262340 × 5376
386FFN_45SwiGLU1 × 262340 × 5376
387Add_45_ffnAdd1 × 262340 × 5376
388RMSNorm_46_1RMSNorm1 × 262340 × 5376
389Attention_46Grouped Query Attn1 × 262340 × 5376
390Add_46_attnAdd1 × 262340 × 5376
391RMSNorm_46_2RMSNorm1 × 262340 × 5376
392FFN_46SwiGLU1 × 262340 × 5376
393Add_46_ffnAdd1 × 262340 × 5376
394RMSNorm_47_1RMSNorm1 × 262340 × 5376
395Attention_47Grouped Query Attn1 × 262340 × 5376
396Add_47_attnAdd1 × 262340 × 5376
397RMSNorm_47_2RMSNorm1 × 262340 × 5376
398FFN_47SwiGLU1 × 262340 × 5376
399Add_47_ffnAdd1 × 262340 × 5376
400RMSNorm_48_1RMSNorm1 × 262340 × 5376
401Attention_48Grouped Query Attn1 × 262340 × 5376
402Add_48_attnAdd1 × 262340 × 5376
403RMSNorm_48_2RMSNorm1 × 262340 × 5376
404FFN_48SwiGLU1 × 262340 × 5376
405Add_48_ffnAdd1 × 262340 × 5376
406RMSNorm_49_1RMSNorm1 × 262340 × 5376
407Attention_49Grouped Query Attn1 × 262340 × 5376
408Add_49_attnAdd1 × 262340 × 5376
409RMSNorm_49_2RMSNorm1 × 262340 × 5376
410FFN_49SwiGLU1 × 262340 × 5376
411Add_49_ffnAdd1 × 262340 × 5376
412RMSNorm_50_1RMSNorm1 × 262340 × 5376
413Attention_50Grouped Query Attn1 × 262340 × 5376
414Add_50_attnAdd1 × 262340 × 5376
415RMSNorm_50_2RMSNorm1 × 262340 × 5376
416FFN_50SwiGLU1 × 262340 × 5376
417Add_50_ffnAdd1 × 262340 × 5376
418RMSNorm_51_1RMSNorm1 × 262340 × 5376
419Attention_51Grouped Query Attn1 × 262340 × 5376
420Add_51_attnAdd1 × 262340 × 5376
421RMSNorm_51_2RMSNorm1 × 262340 × 5376
422FFN_51SwiGLU1 × 262340 × 5376
423Add_51_ffnAdd1 × 262340 × 5376
424RMSNorm_52_1RMSNorm1 × 262340 × 5376
425Attention_52Grouped Query Attn1 × 262340 × 5376
426Add_52_attnAdd1 × 262340 × 5376
427RMSNorm_52_2RMSNorm1 × 262340 × 5376
428FFN_52SwiGLU1 × 262340 × 5376
429Add_52_ffnAdd1 × 262340 × 5376
430RMSNorm_53_1RMSNorm1 × 262340 × 5376
431Attention_53Grouped Query Attn1 × 262340 × 5376
432Add_53_attnAdd1 × 262340 × 5376
433RMSNorm_53_2RMSNorm1 × 262340 × 5376
434FFN_53SwiGLU1 × 262340 × 5376
435Add_53_ffnAdd1 × 262340 × 5376
436RMSNorm_54_1RMSNorm1 × 262340 × 5376
437Attention_54Grouped Query Attn1 × 262340 × 5376
438Add_54_attnAdd1 × 262340 × 5376
439RMSNorm_54_2RMSNorm1 × 262340 × 5376
440FFN_54SwiGLU1 × 262340 × 5376
441Add_54_ffnAdd1 × 262340 × 5376
442RMSNorm_55_1RMSNorm1 × 262340 × 5376
443Attention_55Grouped Query Attn1 × 262340 × 5376
444Add_55_attnAdd1 × 262340 × 5376
445RMSNorm_55_2RMSNorm1 × 262340 × 5376
446FFN_55SwiGLU1 × 262340 × 5376
447Add_55_ffnAdd1 × 262340 × 5376
448RMSNorm_56_1RMSNorm1 × 262340 × 5376
449Attention_56Grouped Query Attn1 × 262340 × 5376
450Add_56_attnAdd1 × 262340 × 5376
451RMSNorm_56_2RMSNorm1 × 262340 × 5376
452FFN_56SwiGLU1 × 262340 × 5376
453Add_56_ffnAdd1 × 262340 × 5376
454RMSNorm_57_1RMSNorm1 × 262340 × 5376
455Attention_57Grouped Query Attn1 × 262340 × 5376
456Add_57_attnAdd1 × 262340 × 5376
457RMSNorm_57_2RMSNorm1 × 262340 × 5376
458FFN_57SwiGLU1 × 262340 × 5376
459Add_57_ffnAdd1 × 262340 × 5376
460RMSNorm_58_1RMSNorm1 × 262340 × 5376
461Attention_58Grouped Query Attn1 × 262340 × 5376
462Add_58_attnAdd1 × 262340 × 5376
463RMSNorm_58_2RMSNorm1 × 262340 × 5376
464FFN_58SwiGLU1 × 262340 × 5376
465Add_58_ffnAdd1 × 262340 × 5376
466RMSNorm_59_1RMSNorm1 × 262340 × 5376
467Attention_59Grouped Query Attn1 × 262340 × 5376
468Add_59_attnAdd1 × 262340 × 5376
469RMSNorm_59_2RMSNorm1 × 262340 × 5376
470FFN_59SwiGLU1 × 262340 × 5376
471Add_59_ffnAdd1 × 262340 × 5376
472RMSNorm_60_1RMSNorm1 × 262340 × 5376
473Attention_60Grouped Query Attn1 × 262340 × 5376
474Add_60_attnAdd1 × 262340 × 5376
475RMSNorm_60_2RMSNorm1 × 262340 × 5376
476FFN_60SwiGLU1 × 262340 × 5376
477Add_60_ffnAdd1 × 262340 × 5376
478OutputOutput1 × 262340 × 5376

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
warnAcross 87 attention layers this design caches 752 KB per token, so a single 8,192-token sequence needs ~6.3 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 farbodtavakkoli/OTel-2.0-LLM-31B-IT --plan --share