ESMC-6B

Reconstructed from its own config.json with no weights read. 2.1M 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

3.36B

3,358,597,120 parameters

In the published checkpoint

6.35B

6,352,005,184 scalars · safetensors.total, read 2026-09-06

Delta

-47.1%

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

322

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$66.36

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

Structure

324 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 × 2560
3	Positional_Embedding	Learned Pos Embed	1 × 512 × 2560
4	Attention_1	Multi-Head Attention	1 × 512 × 2560
5	Add_1	Add	1 × 512 × 2560
6	LayerNorm_1_1	LayerNorm	1 × 512 × 2560
7	FFN_1	Feed Forward	1 × 512 × 2560
8	Attention_2	Multi-Head Attention	1 × 512 × 2560
9	Add_2	Add	1 × 512 × 2560
10	LayerNorm_2_1	LayerNorm	1 × 512 × 2560
11	FFN_2	Feed Forward	1 × 512 × 2560
12	Attention_3	Multi-Head Attention	1 × 512 × 2560
13	Add_3	Add	1 × 512 × 2560
14	LayerNorm_3_1	LayerNorm	1 × 512 × 2560
15	FFN_3	Feed Forward	1 × 512 × 2560
16	Attention_4	Multi-Head Attention	1 × 512 × 2560
17	Add_4	Add	1 × 512 × 2560
18	LayerNorm_4_1	LayerNorm	1 × 512 × 2560
19	FFN_4	Feed Forward	1 × 512 × 2560
20	Attention_5	Multi-Head Attention	1 × 512 × 2560
21	Add_5	Add	1 × 512 × 2560
22	LayerNorm_5_1	LayerNorm	1 × 512 × 2560
23	FFN_5	Feed Forward	1 × 512 × 2560
24	Attention_6	Multi-Head Attention	1 × 512 × 2560
25	Add_6	Add	1 × 512 × 2560
26	LayerNorm_6_1	LayerNorm	1 × 512 × 2560
27	FFN_6	Feed Forward	1 × 512 × 2560
28	Attention_7	Multi-Head Attention	1 × 512 × 2560
29	Add_7	Add	1 × 512 × 2560
30	LayerNorm_7_1	LayerNorm	1 × 512 × 2560
31	FFN_7	Feed Forward	1 × 512 × 2560
32	Attention_8	Multi-Head Attention	1 × 512 × 2560
33	Add_8	Add	1 × 512 × 2560
34	LayerNorm_8_1	LayerNorm	1 × 512 × 2560
35	FFN_8	Feed Forward	1 × 512 × 2560
36	Attention_9	Multi-Head Attention	1 × 512 × 2560
37	Add_9	Add	1 × 512 × 2560
38	LayerNorm_9_1	LayerNorm	1 × 512 × 2560
39	FFN_9	Feed Forward	1 × 512 × 2560
40	Attention_10	Multi-Head Attention	1 × 512 × 2560
41	Add_10	Add	1 × 512 × 2560
42	LayerNorm_10_1	LayerNorm	1 × 512 × 2560
43	FFN_10	Feed Forward	1 × 512 × 2560
44	Attention_11	Multi-Head Attention	1 × 512 × 2560
45	Add_11	Add	1 × 512 × 2560
46	LayerNorm_11_1	LayerNorm	1 × 512 × 2560
47	FFN_11	Feed Forward	1 × 512 × 2560
48	Attention_12	Multi-Head Attention	1 × 512 × 2560
49	Add_12	Add	1 × 512 × 2560
50	LayerNorm_12_1	LayerNorm	1 × 512 × 2560
51	FFN_12	Feed Forward	1 × 512 × 2560
52	Attention_13	Multi-Head Attention	1 × 512 × 2560
53	Add_13	Add	1 × 512 × 2560
54	LayerNorm_13_1	LayerNorm	1 × 512 × 2560
55	FFN_13	Feed Forward	1 × 512 × 2560
56	Attention_14	Multi-Head Attention	1 × 512 × 2560
57	Add_14	Add	1 × 512 × 2560
58	LayerNorm_14_1	LayerNorm	1 × 512 × 2560
59	FFN_14	Feed Forward	1 × 512 × 2560
60	Attention_15	Multi-Head Attention	1 × 512 × 2560
61	Add_15	Add	1 × 512 × 2560
62	LayerNorm_15_1	LayerNorm	1 × 512 × 2560
63	FFN_15	Feed Forward	1 × 512 × 2560
64	Attention_16	Multi-Head Attention	1 × 512 × 2560
65	Add_16	Add	1 × 512 × 2560
66	LayerNorm_16_1	LayerNorm	1 × 512 × 2560
67	FFN_16	Feed Forward	1 × 512 × 2560
68	Attention_17	Multi-Head Attention	1 × 512 × 2560
69	Add_17	Add	1 × 512 × 2560
70	LayerNorm_17_1	LayerNorm	1 × 512 × 2560
71	FFN_17	Feed Forward	1 × 512 × 2560
72	Attention_18	Multi-Head Attention	1 × 512 × 2560
73	Add_18	Add	1 × 512 × 2560
74	LayerNorm_18_1	LayerNorm	1 × 512 × 2560
75	FFN_18	Feed Forward	1 × 512 × 2560
76	Attention_19	Multi-Head Attention	1 × 512 × 2560
77	Add_19	Add	1 × 512 × 2560
78	LayerNorm_19_1	LayerNorm	1 × 512 × 2560
79	FFN_19	Feed Forward	1 × 512 × 2560
80	Attention_20	Multi-Head Attention	1 × 512 × 2560
81	Add_20	Add	1 × 512 × 2560
82	LayerNorm_20_1	LayerNorm	1 × 512 × 2560
83	FFN_20	Feed Forward	1 × 512 × 2560
84	Attention_21	Multi-Head Attention	1 × 512 × 2560
85	Add_21	Add	1 × 512 × 2560
86	LayerNorm_21_1	LayerNorm	1 × 512 × 2560
87	FFN_21	Feed Forward	1 × 512 × 2560
88	Attention_22	Multi-Head Attention	1 × 512 × 2560
89	Add_22	Add	1 × 512 × 2560
90	LayerNorm_22_1	LayerNorm	1 × 512 × 2560
91	FFN_22	Feed Forward	1 × 512 × 2560
92	Attention_23	Multi-Head Attention	1 × 512 × 2560
93	Add_23	Add	1 × 512 × 2560
94	LayerNorm_23_1	LayerNorm	1 × 512 × 2560
95	FFN_23	Feed Forward	1 × 512 × 2560
96	Attention_24	Multi-Head Attention	1 × 512 × 2560
97	Add_24	Add	1 × 512 × 2560
98	LayerNorm_24_1	LayerNorm	1 × 512 × 2560
99	FFN_24	Feed Forward	1 × 512 × 2560
100	Attention_25	Multi-Head Attention	1 × 512 × 2560
101	Add_25	Add	1 × 512 × 2560
102	LayerNorm_25_1	LayerNorm	1 × 512 × 2560
103	FFN_25	Feed Forward	1 × 512 × 2560
104	Attention_26	Multi-Head Attention	1 × 512 × 2560
105	Add_26	Add	1 × 512 × 2560
106	LayerNorm_26_1	LayerNorm	1 × 512 × 2560
107	FFN_26	Feed Forward	1 × 512 × 2560
108	Attention_27	Multi-Head Attention	1 × 512 × 2560
109	Add_27	Add	1 × 512 × 2560
110	LayerNorm_27_1	LayerNorm	1 × 512 × 2560
111	FFN_27	Feed Forward	1 × 512 × 2560
112	Attention_28	Multi-Head Attention	1 × 512 × 2560
113	Add_28	Add	1 × 512 × 2560
114	LayerNorm_28_1	LayerNorm	1 × 512 × 2560
115	FFN_28	Feed Forward	1 × 512 × 2560
116	Attention_29	Multi-Head Attention	1 × 512 × 2560
117	Add_29	Add	1 × 512 × 2560
118	LayerNorm_29_1	LayerNorm	1 × 512 × 2560
119	FFN_29	Feed Forward	1 × 512 × 2560
120	Attention_30	Multi-Head Attention	1 × 512 × 2560
121	Add_30	Add	1 × 512 × 2560
122	LayerNorm_30_1	LayerNorm	1 × 512 × 2560
123	FFN_30	Feed Forward	1 × 512 × 2560
124	Attention_31	Multi-Head Attention	1 × 512 × 2560
125	Add_31	Add	1 × 512 × 2560
126	LayerNorm_31_1	LayerNorm	1 × 512 × 2560
127	FFN_31	Feed Forward	1 × 512 × 2560
128	Attention_32	Multi-Head Attention	1 × 512 × 2560
129	Add_32	Add	1 × 512 × 2560
130	LayerNorm_32_1	LayerNorm	1 × 512 × 2560
131	FFN_32	Feed Forward	1 × 512 × 2560
132	Attention_33	Multi-Head Attention	1 × 512 × 2560
133	Add_33	Add	1 × 512 × 2560
134	LayerNorm_33_1	LayerNorm	1 × 512 × 2560
135	FFN_33	Feed Forward	1 × 512 × 2560
136	Attention_34	Multi-Head Attention	1 × 512 × 2560
137	Add_34	Add	1 × 512 × 2560
138	LayerNorm_34_1	LayerNorm	1 × 512 × 2560
139	FFN_34	Feed Forward	1 × 512 × 2560
140	Attention_35	Multi-Head Attention	1 × 512 × 2560
141	Add_35	Add	1 × 512 × 2560
142	LayerNorm_35_1	LayerNorm	1 × 512 × 2560
143	FFN_35	Feed Forward	1 × 512 × 2560
144	Attention_36	Multi-Head Attention	1 × 512 × 2560
145	Add_36	Add	1 × 512 × 2560
146	LayerNorm_36_1	LayerNorm	1 × 512 × 2560
147	FFN_36	Feed Forward	1 × 512 × 2560
148	Attention_37	Multi-Head Attention	1 × 512 × 2560
149	Add_37	Add	1 × 512 × 2560
150	LayerNorm_37_1	LayerNorm	1 × 512 × 2560
151	FFN_37	Feed Forward	1 × 512 × 2560
152	Attention_38	Multi-Head Attention	1 × 512 × 2560
153	Add_38	Add	1 × 512 × 2560
154	LayerNorm_38_1	LayerNorm	1 × 512 × 2560
155	FFN_38	Feed Forward	1 × 512 × 2560
156	Attention_39	Multi-Head Attention	1 × 512 × 2560
157	Add_39	Add	1 × 512 × 2560
158	LayerNorm_39_1	LayerNorm	1 × 512 × 2560
159	FFN_39	Feed Forward	1 × 512 × 2560
160	Attention_40	Multi-Head Attention	1 × 512 × 2560
161	Add_40	Add	1 × 512 × 2560
162	LayerNorm_40_1	LayerNorm	1 × 512 × 2560
163	FFN_40	Feed Forward	1 × 512 × 2560
164	Attention_41	Multi-Head Attention	1 × 512 × 2560
165	Add_41	Add	1 × 512 × 2560
166	LayerNorm_41_1	LayerNorm	1 × 512 × 2560
167	FFN_41	Feed Forward	1 × 512 × 2560
168	Attention_42	Multi-Head Attention	1 × 512 × 2560
169	Add_42	Add	1 × 512 × 2560
170	LayerNorm_42_1	LayerNorm	1 × 512 × 2560
171	FFN_42	Feed Forward	1 × 512 × 2560
172	Attention_43	Multi-Head Attention	1 × 512 × 2560
173	Add_43	Add	1 × 512 × 2560
174	LayerNorm_43_1	LayerNorm	1 × 512 × 2560
175	FFN_43	Feed Forward	1 × 512 × 2560
176	Attention_44	Multi-Head Attention	1 × 512 × 2560
177	Add_44	Add	1 × 512 × 2560
178	LayerNorm_44_1	LayerNorm	1 × 512 × 2560
179	FFN_44	Feed Forward	1 × 512 × 2560
180	Attention_45	Multi-Head Attention	1 × 512 × 2560
181	Add_45	Add	1 × 512 × 2560
182	LayerNorm_45_1	LayerNorm	1 × 512 × 2560
183	FFN_45	Feed Forward	1 × 512 × 2560
184	Attention_46	Multi-Head Attention	1 × 512 × 2560
185	Add_46	Add	1 × 512 × 2560
186	LayerNorm_46_1	LayerNorm	1 × 512 × 2560
187	FFN_46	Feed Forward	1 × 512 × 2560
188	Attention_47	Multi-Head Attention	1 × 512 × 2560
189	Add_47	Add	1 × 512 × 2560
190	LayerNorm_47_1	LayerNorm	1 × 512 × 2560
191	FFN_47	Feed Forward	1 × 512 × 2560
192	Attention_48	Multi-Head Attention	1 × 512 × 2560
193	Add_48	Add	1 × 512 × 2560
194	LayerNorm_48_1	LayerNorm	1 × 512 × 2560
195	FFN_48	Feed Forward	1 × 512 × 2560
196	Attention_49	Multi-Head Attention	1 × 512 × 2560
197	Add_49	Add	1 × 512 × 2560
198	LayerNorm_49_1	LayerNorm	1 × 512 × 2560
199	FFN_49	Feed Forward	1 × 512 × 2560
200	Attention_50	Multi-Head Attention	1 × 512 × 2560
201	Add_50	Add	1 × 512 × 2560
202	LayerNorm_50_1	LayerNorm	1 × 512 × 2560
203	FFN_50	Feed Forward	1 × 512 × 2560
204	Attention_51	Multi-Head Attention	1 × 512 × 2560
205	Add_51	Add	1 × 512 × 2560
206	LayerNorm_51_1	LayerNorm	1 × 512 × 2560
207	FFN_51	Feed Forward	1 × 512 × 2560
208	Attention_52	Multi-Head Attention	1 × 512 × 2560
209	Add_52	Add	1 × 512 × 2560
210	LayerNorm_52_1	LayerNorm	1 × 512 × 2560
211	FFN_52	Feed Forward	1 × 512 × 2560
212	Attention_53	Multi-Head Attention	1 × 512 × 2560
213	Add_53	Add	1 × 512 × 2560
214	LayerNorm_53_1	LayerNorm	1 × 512 × 2560
215	FFN_53	Feed Forward	1 × 512 × 2560
216	Attention_54	Multi-Head Attention	1 × 512 × 2560
217	Add_54	Add	1 × 512 × 2560
218	LayerNorm_54_1	LayerNorm	1 × 512 × 2560
219	FFN_54	Feed Forward	1 × 512 × 2560
220	Attention_55	Multi-Head Attention	1 × 512 × 2560
221	Add_55	Add	1 × 512 × 2560
222	LayerNorm_55_1	LayerNorm	1 × 512 × 2560
223	FFN_55	Feed Forward	1 × 512 × 2560
224	Attention_56	Multi-Head Attention	1 × 512 × 2560
225	Add_56	Add	1 × 512 × 2560
226	LayerNorm_56_1	LayerNorm	1 × 512 × 2560
227	FFN_56	Feed Forward	1 × 512 × 2560
228	Attention_57	Multi-Head Attention	1 × 512 × 2560
229	Add_57	Add	1 × 512 × 2560
230	LayerNorm_57_1	LayerNorm	1 × 512 × 2560
231	FFN_57	Feed Forward	1 × 512 × 2560
232	Attention_58	Multi-Head Attention	1 × 512 × 2560
233	Add_58	Add	1 × 512 × 2560
234	LayerNorm_58_1	LayerNorm	1 × 512 × 2560
235	FFN_58	Feed Forward	1 × 512 × 2560
236	Attention_59	Multi-Head Attention	1 × 512 × 2560
237	Add_59	Add	1 × 512 × 2560
238	LayerNorm_59_1	LayerNorm	1 × 512 × 2560
239	FFN_59	Feed Forward	1 × 512 × 2560
240	Attention_60	Multi-Head Attention	1 × 512 × 2560
241	Add_60	Add	1 × 512 × 2560
242	LayerNorm_60_1	LayerNorm	1 × 512 × 2560
243	FFN_60	Feed Forward	1 × 512 × 2560
244	Attention_61	Multi-Head Attention	1 × 512 × 2560
245	Add_61	Add	1 × 512 × 2560
246	LayerNorm_61_1	LayerNorm	1 × 512 × 2560
247	FFN_61	Feed Forward	1 × 512 × 2560
248	Attention_62	Multi-Head Attention	1 × 512 × 2560
249	Add_62	Add	1 × 512 × 2560
250	LayerNorm_62_1	LayerNorm	1 × 512 × 2560
251	FFN_62	Feed Forward	1 × 512 × 2560
252	Attention_63	Multi-Head Attention	1 × 512 × 2560
253	Add_63	Add	1 × 512 × 2560
254	LayerNorm_63_1	LayerNorm	1 × 512 × 2560
255	FFN_63	Feed Forward	1 × 512 × 2560
256	Attention_64	Multi-Head Attention	1 × 512 × 2560
257	Add_64	Add	1 × 512 × 2560
258	LayerNorm_64_1	LayerNorm	1 × 512 × 2560
259	FFN_64	Feed Forward	1 × 512 × 2560
260	Attention_65	Multi-Head Attention	1 × 512 × 2560
261	Add_65	Add	1 × 512 × 2560
262	LayerNorm_65_1	LayerNorm	1 × 512 × 2560
263	FFN_65	Feed Forward	1 × 512 × 2560
264	Attention_66	Multi-Head Attention	1 × 512 × 2560
265	Add_66	Add	1 × 512 × 2560
266	LayerNorm_66_1	LayerNorm	1 × 512 × 2560
267	FFN_66	Feed Forward	1 × 512 × 2560
268	Attention_67	Multi-Head Attention	1 × 512 × 2560
269	Add_67	Add	1 × 512 × 2560
270	LayerNorm_67_1	LayerNorm	1 × 512 × 2560
271	FFN_67	Feed Forward	1 × 512 × 2560
272	Attention_68	Multi-Head Attention	1 × 512 × 2560
273	Add_68	Add	1 × 512 × 2560
274	LayerNorm_68_1	LayerNorm	1 × 512 × 2560
275	FFN_68	Feed Forward	1 × 512 × 2560
276	Attention_69	Multi-Head Attention	1 × 512 × 2560
277	Add_69	Add	1 × 512 × 2560
278	LayerNorm_69_1	LayerNorm	1 × 512 × 2560
279	FFN_69	Feed Forward	1 × 512 × 2560
280	Attention_70	Multi-Head Attention	1 × 512 × 2560
281	Add_70	Add	1 × 512 × 2560
282	LayerNorm_70_1	LayerNorm	1 × 512 × 2560
283	FFN_70	Feed Forward	1 × 512 × 2560
284	Attention_71	Multi-Head Attention	1 × 512 × 2560
285	Add_71	Add	1 × 512 × 2560
286	LayerNorm_71_1	LayerNorm	1 × 512 × 2560
287	FFN_71	Feed Forward	1 × 512 × 2560
288	Attention_72	Multi-Head Attention	1 × 512 × 2560
289	Add_72	Add	1 × 512 × 2560
290	LayerNorm_72_1	LayerNorm	1 × 512 × 2560
291	FFN_72	Feed Forward	1 × 512 × 2560
292	Attention_73	Multi-Head Attention	1 × 512 × 2560
293	Add_73	Add	1 × 512 × 2560
294	LayerNorm_73_1	LayerNorm	1 × 512 × 2560
295	FFN_73	Feed Forward	1 × 512 × 2560
296	Attention_74	Multi-Head Attention	1 × 512 × 2560
297	Add_74	Add	1 × 512 × 2560
298	LayerNorm_74_1	LayerNorm	1 × 512 × 2560
299	FFN_74	Feed Forward	1 × 512 × 2560
300	Attention_75	Multi-Head Attention	1 × 512 × 2560
301	Add_75	Add	1 × 512 × 2560
302	LayerNorm_75_1	LayerNorm	1 × 512 × 2560
303	FFN_75	Feed Forward	1 × 512 × 2560
304	Attention_76	Multi-Head Attention	1 × 512 × 2560
305	Add_76	Add	1 × 512 × 2560
306	LayerNorm_76_1	LayerNorm	1 × 512 × 2560
307	FFN_76	Feed Forward	1 × 512 × 2560
308	Attention_77	Multi-Head Attention	1 × 512 × 2560
309	Add_77	Add	1 × 512 × 2560
310	LayerNorm_77_1	LayerNorm	1 × 512 × 2560
311	FFN_77	Feed Forward	1 × 512 × 2560
312	Attention_78	Multi-Head Attention	1 × 512 × 2560
313	Add_78	Add	1 × 512 × 2560
314	LayerNorm_78_1	LayerNorm	1 × 512 × 2560
315	FFN_78	Feed Forward	1 × 512 × 2560
316	Attention_79	Multi-Head Attention	1 × 512 × 2560
317	Add_79	Add	1 × 512 × 2560
318	LayerNorm_79_1	LayerNorm	1 × 512 × 2560
319	FFN_79	Feed Forward	1 × 512 × 2560
320	Attention_80	Multi-Head Attention	1 × 512 × 2560
321	Add_80	Add	1 × 512 × 2560
322	LayerNorm_80_1	LayerNorm	1 × 512 × 2560
323	FFN_80	Feed Forward	1 × 512 × 2560
324	Output	Output	1 × 512 × 2560

What the verifier says

info80 attention layers at embedDim 2560 cache full per-head K/V: about 800 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 800 KB per token, so a single 8,192-token sequence needs ~6.7 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 biohub/ESMC-6B --plan --share

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