bloomz-560m

Reconstructed from its own config.json with no weights read. 766K 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

364M

363,571,200 parameters

In the published checkpoint

559M

559,214,592 scalars · safetensors.total, read 2023-05-27

Delta

-35.0%

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

146

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$3.48

Card	Memory
T4 (16GB)	weights + activations	fits
A100 (40GB)	weights + activations	fits
H100 (80GB)	weights + activations	fits

Structure

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

	Layer	Type	Output shape
1	Input	Input	1 × 512
2	Embedding	Embedding	1 × 512 × 768
3	Positional_Embedding	Learned Pos Embed	1 × 512 × 768
4	LayerNorm_1_1	LayerNorm	1 × 512 × 768
5	Attention_1	Multi-Head Attention	1 × 512 × 768
6	Add_1_attn	Add	1 × 512 × 768
7	LayerNorm_1_2	LayerNorm	1 × 512 × 768
8	FFN_1	Feed Forward	1 × 512 × 768
9	Add_1_ffn	Add	1 × 512 × 768
10	LayerNorm_2_1	LayerNorm	1 × 512 × 768
11	Attention_2	Multi-Head Attention	1 × 512 × 768
12	Add_2_attn	Add	1 × 512 × 768
13	LayerNorm_2_2	LayerNorm	1 × 512 × 768
14	FFN_2	Feed Forward	1 × 512 × 768
15	Add_2_ffn	Add	1 × 512 × 768
16	LayerNorm_3_1	LayerNorm	1 × 512 × 768
17	Attention_3	Multi-Head Attention	1 × 512 × 768
18	Add_3_attn	Add	1 × 512 × 768
19	LayerNorm_3_2	LayerNorm	1 × 512 × 768
20	FFN_3	Feed Forward	1 × 512 × 768
21	Add_3_ffn	Add	1 × 512 × 768
22	LayerNorm_4_1	LayerNorm	1 × 512 × 768
23	Attention_4	Multi-Head Attention	1 × 512 × 768
24	Add_4_attn	Add	1 × 512 × 768
25	LayerNorm_4_2	LayerNorm	1 × 512 × 768
26	FFN_4	Feed Forward	1 × 512 × 768
27	Add_4_ffn	Add	1 × 512 × 768
28	LayerNorm_5_1	LayerNorm	1 × 512 × 768
29	Attention_5	Multi-Head Attention	1 × 512 × 768
30	Add_5_attn	Add	1 × 512 × 768
31	LayerNorm_5_2	LayerNorm	1 × 512 × 768
32	FFN_5	Feed Forward	1 × 512 × 768
33	Add_5_ffn	Add	1 × 512 × 768
34	LayerNorm_6_1	LayerNorm	1 × 512 × 768
35	Attention_6	Multi-Head Attention	1 × 512 × 768
36	Add_6_attn	Add	1 × 512 × 768
37	LayerNorm_6_2	LayerNorm	1 × 512 × 768
38	FFN_6	Feed Forward	1 × 512 × 768
39	Add_6_ffn	Add	1 × 512 × 768
40	LayerNorm_7_1	LayerNorm	1 × 512 × 768
41	Attention_7	Multi-Head Attention	1 × 512 × 768
42	Add_7_attn	Add	1 × 512 × 768
43	LayerNorm_7_2	LayerNorm	1 × 512 × 768
44	FFN_7	Feed Forward	1 × 512 × 768
45	Add_7_ffn	Add	1 × 512 × 768
46	LayerNorm_8_1	LayerNorm	1 × 512 × 768
47	Attention_8	Multi-Head Attention	1 × 512 × 768
48	Add_8_attn	Add	1 × 512 × 768
49	LayerNorm_8_2	LayerNorm	1 × 512 × 768
50	FFN_8	Feed Forward	1 × 512 × 768
51	Add_8_ffn	Add	1 × 512 × 768
52	LayerNorm_9_1	LayerNorm	1 × 512 × 768
53	Attention_9	Multi-Head Attention	1 × 512 × 768
54	Add_9_attn	Add	1 × 512 × 768
55	LayerNorm_9_2	LayerNorm	1 × 512 × 768
56	FFN_9	Feed Forward	1 × 512 × 768
57	Add_9_ffn	Add	1 × 512 × 768
58	LayerNorm_10_1	LayerNorm	1 × 512 × 768
59	Attention_10	Multi-Head Attention	1 × 512 × 768
60	Add_10_attn	Add	1 × 512 × 768
61	LayerNorm_10_2	LayerNorm	1 × 512 × 768
62	FFN_10	Feed Forward	1 × 512 × 768
63	Add_10_ffn	Add	1 × 512 × 768
64	LayerNorm_11_1	LayerNorm	1 × 512 × 768
65	Attention_11	Multi-Head Attention	1 × 512 × 768
66	Add_11_attn	Add	1 × 512 × 768
67	LayerNorm_11_2	LayerNorm	1 × 512 × 768
68	FFN_11	Feed Forward	1 × 512 × 768
69	Add_11_ffn	Add	1 × 512 × 768
70	LayerNorm_12_1	LayerNorm	1 × 512 × 768
71	Attention_12	Multi-Head Attention	1 × 512 × 768
72	Add_12_attn	Add	1 × 512 × 768
73	LayerNorm_12_2	LayerNorm	1 × 512 × 768
74	FFN_12	Feed Forward	1 × 512 × 768
75	Add_12_ffn	Add	1 × 512 × 768
76	LayerNorm_13_1	LayerNorm	1 × 512 × 768
77	Attention_13	Multi-Head Attention	1 × 512 × 768
78	Add_13_attn	Add	1 × 512 × 768
79	LayerNorm_13_2	LayerNorm	1 × 512 × 768
80	FFN_13	Feed Forward	1 × 512 × 768
81	Add_13_ffn	Add	1 × 512 × 768
82	LayerNorm_14_1	LayerNorm	1 × 512 × 768
83	Attention_14	Multi-Head Attention	1 × 512 × 768
84	Add_14_attn	Add	1 × 512 × 768
85	LayerNorm_14_2	LayerNorm	1 × 512 × 768
86	FFN_14	Feed Forward	1 × 512 × 768
87	Add_14_ffn	Add	1 × 512 × 768
88	LayerNorm_15_1	LayerNorm	1 × 512 × 768
89	Attention_15	Multi-Head Attention	1 × 512 × 768
90	Add_15_attn	Add	1 × 512 × 768
91	LayerNorm_15_2	LayerNorm	1 × 512 × 768
92	FFN_15	Feed Forward	1 × 512 × 768
93	Add_15_ffn	Add	1 × 512 × 768
94	LayerNorm_16_1	LayerNorm	1 × 512 × 768
95	Attention_16	Multi-Head Attention	1 × 512 × 768
96	Add_16_attn	Add	1 × 512 × 768
97	LayerNorm_16_2	LayerNorm	1 × 512 × 768
98	FFN_16	Feed Forward	1 × 512 × 768
99	Add_16_ffn	Add	1 × 512 × 768
100	LayerNorm_17_1	LayerNorm	1 × 512 × 768
101	Attention_17	Multi-Head Attention	1 × 512 × 768
102	Add_17_attn	Add	1 × 512 × 768
103	LayerNorm_17_2	LayerNorm	1 × 512 × 768
104	FFN_17	Feed Forward	1 × 512 × 768
105	Add_17_ffn	Add	1 × 512 × 768
106	LayerNorm_18_1	LayerNorm	1 × 512 × 768
107	Attention_18	Multi-Head Attention	1 × 512 × 768
108	Add_18_attn	Add	1 × 512 × 768
109	LayerNorm_18_2	LayerNorm	1 × 512 × 768
110	FFN_18	Feed Forward	1 × 512 × 768
111	Add_18_ffn	Add	1 × 512 × 768
112	LayerNorm_19_1	LayerNorm	1 × 512 × 768
113	Attention_19	Multi-Head Attention	1 × 512 × 768
114	Add_19_attn	Add	1 × 512 × 768
115	LayerNorm_19_2	LayerNorm	1 × 512 × 768
116	FFN_19	Feed Forward	1 × 512 × 768
117	Add_19_ffn	Add	1 × 512 × 768
118	LayerNorm_20_1	LayerNorm	1 × 512 × 768
119	Attention_20	Multi-Head Attention	1 × 512 × 768
120	Add_20_attn	Add	1 × 512 × 768
121	LayerNorm_20_2	LayerNorm	1 × 512 × 768
122	FFN_20	Feed Forward	1 × 512 × 768
123	Add_20_ffn	Add	1 × 512 × 768
124	LayerNorm_21_1	LayerNorm	1 × 512 × 768
125	Attention_21	Multi-Head Attention	1 × 512 × 768
126	Add_21_attn	Add	1 × 512 × 768
127	LayerNorm_21_2	LayerNorm	1 × 512 × 768
128	FFN_21	Feed Forward	1 × 512 × 768
129	Add_21_ffn	Add	1 × 512 × 768
130	LayerNorm_22_1	LayerNorm	1 × 512 × 768
131	Attention_22	Multi-Head Attention	1 × 512 × 768
132	Add_22_attn	Add	1 × 512 × 768
133	LayerNorm_22_2	LayerNorm	1 × 512 × 768
134	FFN_22	Feed Forward	1 × 512 × 768
135	Add_22_ffn	Add	1 × 512 × 768
136	LayerNorm_23_1	LayerNorm	1 × 512 × 768
137	Attention_23	Multi-Head Attention	1 × 512 × 768
138	Add_23_attn	Add	1 × 512 × 768
139	LayerNorm_23_2	LayerNorm	1 × 512 × 768
140	FFN_23	Feed Forward	1 × 512 × 768
141	Add_23_ffn	Add	1 × 512 × 768
142	LayerNorm_24_1	LayerNorm	1 × 512 × 768
143	Attention_24	Multi-Head Attention	1 × 512 × 768
144	Add_24_attn	Add	1 × 512 × 768
145	LayerNorm_24_2	LayerNorm	1 × 512 × 768
146	FFN_24	Feed Forward	1 × 512 × 768
147	Add_24_ffn	Add	1 × 512 × 768
148	Output	Output	1 × 512 × 768

What the verifier says

infoAt 24 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 bigscience/bloomz-560m --plan --share

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