Models / internvl_chat

InternVL2-1B

Reconstructed from its own config.json with no weights read. 595K 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

935M

934,619,239 parameters

In the published checkpoint

938M

938,158,976 scalars · safetensors.total, read 2025-03-25

Delta

-0.38%

custom-code This repository ships its own modeling code (`auto_map`, e.g. `configuration_internvl_chat.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

249

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$1556.31

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

Structure

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

	Layer	Type	Output shape
1	Input	Input	1 × 32768
2	Embedding	Embedding	1 × 32768 × 896
3	RoPE	RoPE	1 × 32768 × 896
4	Vision input	Input	3 × 448 × 448
5	PatchEmbed	Patch Embed	1024 × 1024
6	Patch_Position_Embedding	Learned Pos Embed	1024 × 1024
7	Vision_LN_1	LayerNorm	1024 × 1024
8	Vision_Attn_1	Multi-Head Attention	1024 × 1024
9	Vision_Add_1	Add	1024 × 1024
10	Vision_FFN_1	Feed Forward	1024 × 1024
11	Vision_LN_2	LayerNorm	1024 × 1024
12	Vision_Attn_2	Multi-Head Attention	1024 × 1024
13	Vision_Add_2	Add	1024 × 1024
14	Vision_FFN_2	Feed Forward	1024 × 1024
15	Vision_LN_3	LayerNorm	1024 × 1024
16	Vision_Attn_3	Multi-Head Attention	1024 × 1024
17	Vision_Add_3	Add	1024 × 1024
18	Vision_FFN_3	Feed Forward	1024 × 1024
19	Vision_LN_4	LayerNorm	1024 × 1024
20	Vision_Attn_4	Multi-Head Attention	1024 × 1024
21	Vision_Add_4	Add	1024 × 1024
22	Vision_FFN_4	Feed Forward	1024 × 1024
23	Vision_LN_5	LayerNorm	1024 × 1024
24	Vision_Attn_5	Multi-Head Attention	1024 × 1024
25	Vision_Add_5	Add	1024 × 1024
26	Vision_FFN_5	Feed Forward	1024 × 1024
27	Vision_LN_6	LayerNorm	1024 × 1024
28	Vision_Attn_6	Multi-Head Attention	1024 × 1024
29	Vision_Add_6	Add	1024 × 1024
30	Vision_FFN_6	Feed Forward	1024 × 1024
31	Vision_LN_7	LayerNorm	1024 × 1024
32	Vision_Attn_7	Multi-Head Attention	1024 × 1024
33	Vision_Add_7	Add	1024 × 1024
34	Vision_FFN_7	Feed Forward	1024 × 1024
35	Vision_LN_8	LayerNorm	1024 × 1024
36	Vision_Attn_8	Multi-Head Attention	1024 × 1024
37	Vision_Add_8	Add	1024 × 1024
38	Vision_FFN_8	Feed Forward	1024 × 1024
39	Vision_LN_9	LayerNorm	1024 × 1024
40	Vision_Attn_9	Multi-Head Attention	1024 × 1024
41	Vision_Add_9	Add	1024 × 1024
42	Vision_FFN_9	Feed Forward	1024 × 1024
43	Vision_LN_10	LayerNorm	1024 × 1024
44	Vision_Attn_10	Multi-Head Attention	1024 × 1024
45	Vision_Add_10	Add	1024 × 1024
46	Vision_FFN_10	Feed Forward	1024 × 1024
47	Vision_LN_11	LayerNorm	1024 × 1024
48	Vision_Attn_11	Multi-Head Attention	1024 × 1024
49	Vision_Add_11	Add	1024 × 1024
50	Vision_FFN_11	Feed Forward	1024 × 1024
51	Vision_LN_12	LayerNorm	1024 × 1024
52	Vision_Attn_12	Multi-Head Attention	1024 × 1024
53	Vision_Add_12	Add	1024 × 1024
54	Vision_FFN_12	Feed Forward	1024 × 1024
55	Vision_LN_13	LayerNorm	1024 × 1024
56	Vision_Attn_13	Multi-Head Attention	1024 × 1024
57	Vision_Add_13	Add	1024 × 1024
58	Vision_FFN_13	Feed Forward	1024 × 1024
59	Vision_LN_14	LayerNorm	1024 × 1024
60	Vision_Attn_14	Multi-Head Attention	1024 × 1024
61	Vision_Add_14	Add	1024 × 1024
62	Vision_FFN_14	Feed Forward	1024 × 1024
63	Vision_LN_15	LayerNorm	1024 × 1024
64	Vision_Attn_15	Multi-Head Attention	1024 × 1024
65	Vision_Add_15	Add	1024 × 1024
66	Vision_FFN_15	Feed Forward	1024 × 1024
67	Vision_LN_16	LayerNorm	1024 × 1024
68	Vision_Attn_16	Multi-Head Attention	1024 × 1024
69	Vision_Add_16	Add	1024 × 1024
70	Vision_FFN_16	Feed Forward	1024 × 1024
71	Vision_LN_17	LayerNorm	1024 × 1024
72	Vision_Attn_17	Multi-Head Attention	1024 × 1024
73	Vision_Add_17	Add	1024 × 1024
74	Vision_FFN_17	Feed Forward	1024 × 1024
75	Vision_LN_18	LayerNorm	1024 × 1024
76	Vision_Attn_18	Multi-Head Attention	1024 × 1024
77	Vision_Add_18	Add	1024 × 1024
78	Vision_FFN_18	Feed Forward	1024 × 1024
79	Vision_LN_19	LayerNorm	1024 × 1024
80	Vision_Attn_19	Multi-Head Attention	1024 × 1024
81	Vision_Add_19	Add	1024 × 1024
82	Vision_FFN_19	Feed Forward	1024 × 1024
83	Vision_LN_20	LayerNorm	1024 × 1024
84	Vision_Attn_20	Multi-Head Attention	1024 × 1024
85	Vision_Add_20	Add	1024 × 1024
86	Vision_FFN_20	Feed Forward	1024 × 1024
87	Vision_LN_21	LayerNorm	1024 × 1024
88	Vision_Attn_21	Multi-Head Attention	1024 × 1024
89	Vision_Add_21	Add	1024 × 1024
90	Vision_FFN_21	Feed Forward	1024 × 1024
91	Vision_LN_22	LayerNorm	1024 × 1024
92	Vision_Attn_22	Multi-Head Attention	1024 × 1024
93	Vision_Add_22	Add	1024 × 1024
94	Vision_FFN_22	Feed Forward	1024 × 1024
95	Vision_LN_23	LayerNorm	1024 × 1024
96	Vision_Attn_23	Multi-Head Attention	1024 × 1024
97	Vision_Add_23	Add	1024 × 1024
98	Vision_FFN_23	Feed Forward	1024 × 1024
99	Vision_LN_24	LayerNorm	1024 × 1024
100	Vision_Attn_24	Multi-Head Attention	1024 × 1024
101	Vision_Add_24	Add	1024 × 1024
102	Vision_FFN_24	Feed Forward	1024 × 1024
103	Vision projector	Projection	1024 × 896
104	Vision tokens	Reshape	1 × 1024 × 896
105	Multimodal fusion (concat tokens)	Concatenate	1 × 33792 × 896
106	RMSNorm_1_1	RMSNorm	1 × 33792 × 896
107	Attention_1	Grouped Query Attn	1 × 33792 × 896
108	Add_1_attn	Add	1 × 33792 × 896
109	RMSNorm_1_2	RMSNorm	1 × 33792 × 896
110	FFN_1	SwiGLU	1 × 33792 × 896
111	Add_1_ffn	Add	1 × 33792 × 896
112	RMSNorm_2_1	RMSNorm	1 × 33792 × 896
113	Attention_2	Grouped Query Attn	1 × 33792 × 896
114	Add_2_attn	Add	1 × 33792 × 896
115	RMSNorm_2_2	RMSNorm	1 × 33792 × 896
116	FFN_2	SwiGLU	1 × 33792 × 896
117	Add_2_ffn	Add	1 × 33792 × 896
118	RMSNorm_3_1	RMSNorm	1 × 33792 × 896
119	Attention_3	Grouped Query Attn	1 × 33792 × 896
120	Add_3_attn	Add	1 × 33792 × 896
121	RMSNorm_3_2	RMSNorm	1 × 33792 × 896
122	FFN_3	SwiGLU	1 × 33792 × 896
123	Add_3_ffn	Add	1 × 33792 × 896
124	RMSNorm_4_1	RMSNorm	1 × 33792 × 896
125	Attention_4	Grouped Query Attn	1 × 33792 × 896
126	Add_4_attn	Add	1 × 33792 × 896
127	RMSNorm_4_2	RMSNorm	1 × 33792 × 896
128	FFN_4	SwiGLU	1 × 33792 × 896
129	Add_4_ffn	Add	1 × 33792 × 896
130	RMSNorm_5_1	RMSNorm	1 × 33792 × 896
131	Attention_5	Grouped Query Attn	1 × 33792 × 896
132	Add_5_attn	Add	1 × 33792 × 896
133	RMSNorm_5_2	RMSNorm	1 × 33792 × 896
134	FFN_5	SwiGLU	1 × 33792 × 896
135	Add_5_ffn	Add	1 × 33792 × 896
136	RMSNorm_6_1	RMSNorm	1 × 33792 × 896
137	Attention_6	Grouped Query Attn	1 × 33792 × 896
138	Add_6_attn	Add	1 × 33792 × 896
139	RMSNorm_6_2	RMSNorm	1 × 33792 × 896
140	FFN_6	SwiGLU	1 × 33792 × 896
141	Add_6_ffn	Add	1 × 33792 × 896
142	RMSNorm_7_1	RMSNorm	1 × 33792 × 896
143	Attention_7	Grouped Query Attn	1 × 33792 × 896
144	Add_7_attn	Add	1 × 33792 × 896
145	RMSNorm_7_2	RMSNorm	1 × 33792 × 896
146	FFN_7	SwiGLU	1 × 33792 × 896
147	Add_7_ffn	Add	1 × 33792 × 896
148	RMSNorm_8_1	RMSNorm	1 × 33792 × 896
149	Attention_8	Grouped Query Attn	1 × 33792 × 896
150	Add_8_attn	Add	1 × 33792 × 896
151	RMSNorm_8_2	RMSNorm	1 × 33792 × 896
152	FFN_8	SwiGLU	1 × 33792 × 896
153	Add_8_ffn	Add	1 × 33792 × 896
154	RMSNorm_9_1	RMSNorm	1 × 33792 × 896
155	Attention_9	Grouped Query Attn	1 × 33792 × 896
156	Add_9_attn	Add	1 × 33792 × 896
157	RMSNorm_9_2	RMSNorm	1 × 33792 × 896
158	FFN_9	SwiGLU	1 × 33792 × 896
159	Add_9_ffn	Add	1 × 33792 × 896
160	RMSNorm_10_1	RMSNorm	1 × 33792 × 896
161	Attention_10	Grouped Query Attn	1 × 33792 × 896
162	Add_10_attn	Add	1 × 33792 × 896
163	RMSNorm_10_2	RMSNorm	1 × 33792 × 896
164	FFN_10	SwiGLU	1 × 33792 × 896
165	Add_10_ffn	Add	1 × 33792 × 896
166	RMSNorm_11_1	RMSNorm	1 × 33792 × 896
167	Attention_11	Grouped Query Attn	1 × 33792 × 896
168	Add_11_attn	Add	1 × 33792 × 896
169	RMSNorm_11_2	RMSNorm	1 × 33792 × 896
170	FFN_11	SwiGLU	1 × 33792 × 896
171	Add_11_ffn	Add	1 × 33792 × 896
172	RMSNorm_12_1	RMSNorm	1 × 33792 × 896
173	Attention_12	Grouped Query Attn	1 × 33792 × 896
174	Add_12_attn	Add	1 × 33792 × 896
175	RMSNorm_12_2	RMSNorm	1 × 33792 × 896
176	FFN_12	SwiGLU	1 × 33792 × 896
177	Add_12_ffn	Add	1 × 33792 × 896
178	RMSNorm_13_1	RMSNorm	1 × 33792 × 896
179	Attention_13	Grouped Query Attn	1 × 33792 × 896
180	Add_13_attn	Add	1 × 33792 × 896
181	RMSNorm_13_2	RMSNorm	1 × 33792 × 896
182	FFN_13	SwiGLU	1 × 33792 × 896
183	Add_13_ffn	Add	1 × 33792 × 896
184	RMSNorm_14_1	RMSNorm	1 × 33792 × 896
185	Attention_14	Grouped Query Attn	1 × 33792 × 896
186	Add_14_attn	Add	1 × 33792 × 896
187	RMSNorm_14_2	RMSNorm	1 × 33792 × 896
188	FFN_14	SwiGLU	1 × 33792 × 896
189	Add_14_ffn	Add	1 × 33792 × 896
190	RMSNorm_15_1	RMSNorm	1 × 33792 × 896
191	Attention_15	Grouped Query Attn	1 × 33792 × 896
192	Add_15_attn	Add	1 × 33792 × 896
193	RMSNorm_15_2	RMSNorm	1 × 33792 × 896
194	FFN_15	SwiGLU	1 × 33792 × 896
195	Add_15_ffn	Add	1 × 33792 × 896
196	RMSNorm_16_1	RMSNorm	1 × 33792 × 896
197	Attention_16	Grouped Query Attn	1 × 33792 × 896
198	Add_16_attn	Add	1 × 33792 × 896
199	RMSNorm_16_2	RMSNorm	1 × 33792 × 896
200	FFN_16	SwiGLU	1 × 33792 × 896
201	Add_16_ffn	Add	1 × 33792 × 896
202	RMSNorm_17_1	RMSNorm	1 × 33792 × 896
203	Attention_17	Grouped Query Attn	1 × 33792 × 896
204	Add_17_attn	Add	1 × 33792 × 896
205	RMSNorm_17_2	RMSNorm	1 × 33792 × 896
206	FFN_17	SwiGLU	1 × 33792 × 896
207	Add_17_ffn	Add	1 × 33792 × 896
208	RMSNorm_18_1	RMSNorm	1 × 33792 × 896
209	Attention_18	Grouped Query Attn	1 × 33792 × 896
210	Add_18_attn	Add	1 × 33792 × 896
211	RMSNorm_18_2	RMSNorm	1 × 33792 × 896
212	FFN_18	SwiGLU	1 × 33792 × 896
213	Add_18_ffn	Add	1 × 33792 × 896
214	RMSNorm_19_1	RMSNorm	1 × 33792 × 896
215	Attention_19	Grouped Query Attn	1 × 33792 × 896
216	Add_19_attn	Add	1 × 33792 × 896
217	RMSNorm_19_2	RMSNorm	1 × 33792 × 896
218	FFN_19	SwiGLU	1 × 33792 × 896
219	Add_19_ffn	Add	1 × 33792 × 896
220	RMSNorm_20_1	RMSNorm	1 × 33792 × 896
221	Attention_20	Grouped Query Attn	1 × 33792 × 896
222	Add_20_attn	Add	1 × 33792 × 896
223	RMSNorm_20_2	RMSNorm	1 × 33792 × 896
224	FFN_20	SwiGLU	1 × 33792 × 896
225	Add_20_ffn	Add	1 × 33792 × 896
226	RMSNorm_21_1	RMSNorm	1 × 33792 × 896
227	Attention_21	Grouped Query Attn	1 × 33792 × 896
228	Add_21_attn	Add	1 × 33792 × 896
229	RMSNorm_21_2	RMSNorm	1 × 33792 × 896
230	FFN_21	SwiGLU	1 × 33792 × 896
231	Add_21_ffn	Add	1 × 33792 × 896
232	RMSNorm_22_1	RMSNorm	1 × 33792 × 896
233	Attention_22	Grouped Query Attn	1 × 33792 × 896
234	Add_22_attn	Add	1 × 33792 × 896
235	RMSNorm_22_2	RMSNorm	1 × 33792 × 896
236	FFN_22	SwiGLU	1 × 33792 × 896
237	Add_22_ffn	Add	1 × 33792 × 896
238	RMSNorm_23_1	RMSNorm	1 × 33792 × 896
239	Attention_23	Grouped Query Attn	1 × 33792 × 896
240	Add_23_attn	Add	1 × 33792 × 896
241	RMSNorm_23_2	RMSNorm	1 × 33792 × 896
242	FFN_23	SwiGLU	1 × 33792 × 896
243	Add_23_ffn	Add	1 × 33792 × 896
244	RMSNorm_24_1	RMSNorm	1 × 33792 × 896
245	Attention_24	Grouped Query Attn	1 × 33792 × 896
246	Add_24_attn	Add	1 × 33792 × 896
247	RMSNorm_24_2	RMSNorm	1 × 33792 × 896
248	FFN_24	SwiGLU	1 × 33792 × 896
249	Add_24_ffn	Add	1 × 33792 × 896
250	Final_RMSNorm	RMSNorm	1 × 33792 × 896
251	LM_Head	Linear	1 × 33792 × 151655
252	Output	Output	1 × 33792 × 151655

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

infoLLaMA uses intermediateSize ≈ ⌊(8/3 × D) / 256⌋ × 256. Current: 4864 (5.43× embedDim). Expected: ~2304. Fix: Set intermediateSize to 2304 for embedDim=896.
swiglu-dim-convention

infoAt 48 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 OpenGVLab/InternVL2-1B --plan --share

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

Other internvl_chat checkpoints

InternVL2-26B

25.42B derived · -0.38% against the checkpoint

InternVL2-2B

2.20B derived · -0.48% against the checkpoint

InternVL2_5-4B

3.70B derived · -0.28% against the checkpoint

InternVL3-1B

935M derived · -0.38% against the checkpoint