Models / kimi_k25

Kimi-K2.6

Reconstructed from its own config.json with no weights read. 562K 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

31.71B

31,713,123,328 parameters

In the published checkpoint

1027B

1,026,879,376,368 scalars · safetensors.total, read 2026-05-19

Delta

-96.9%

quantized This checkpoint is stored quantized (compressed-tensors). The tensor count in the file counts stored elements under a packing scheme, not logical parameters, so the two numbers below are not measuring the same thing in either direction.

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

246

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$1405738.89

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

Structure

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

	Layer	Type	Output shape
1	Input	Input	1 × 262144
2	Embedding	Embedding	1 × 262144 × 7168
3	Positional_Embedding	Learned Pos Embed	1 × 262144 × 7168
4	Attention_1	Multi-Head Attention	1 × 262144 × 7168
5	Add_1	Add	1 × 262144 × 7168
6	LayerNorm_1_1	LayerNorm	1 × 262144 × 7168
7	FFN_1	Feed Forward	1 × 262144 × 7168
8	Attention_2	Multi-Head Attention	1 × 262144 × 7168
9	Add_2	Add	1 × 262144 × 7168
10	LayerNorm_2_1	LayerNorm	1 × 262144 × 7168
11	FFN_2	Feed Forward	1 × 262144 × 7168
12	Attention_3	Multi-Head Attention	1 × 262144 × 7168
13	Add_3	Add	1 × 262144 × 7168
14	LayerNorm_3_1	LayerNorm	1 × 262144 × 7168
15	FFN_3	Feed Forward	1 × 262144 × 7168
16	Attention_4	Multi-Head Attention	1 × 262144 × 7168
17	Add_4	Add	1 × 262144 × 7168
18	LayerNorm_4_1	LayerNorm	1 × 262144 × 7168
19	FFN_4	Feed Forward	1 × 262144 × 7168
20	Attention_5	Multi-Head Attention	1 × 262144 × 7168
21	Add_5	Add	1 × 262144 × 7168
22	LayerNorm_5_1	LayerNorm	1 × 262144 × 7168
23	FFN_5	Feed Forward	1 × 262144 × 7168
24	Attention_6	Multi-Head Attention	1 × 262144 × 7168
25	Add_6	Add	1 × 262144 × 7168
26	LayerNorm_6_1	LayerNorm	1 × 262144 × 7168
27	FFN_6	Feed Forward	1 × 262144 × 7168
28	Attention_7	Multi-Head Attention	1 × 262144 × 7168
29	Add_7	Add	1 × 262144 × 7168
30	LayerNorm_7_1	LayerNorm	1 × 262144 × 7168
31	FFN_7	Feed Forward	1 × 262144 × 7168
32	Attention_8	Multi-Head Attention	1 × 262144 × 7168
33	Add_8	Add	1 × 262144 × 7168
34	LayerNorm_8_1	LayerNorm	1 × 262144 × 7168
35	FFN_8	Feed Forward	1 × 262144 × 7168
36	Attention_9	Multi-Head Attention	1 × 262144 × 7168
37	Add_9	Add	1 × 262144 × 7168
38	LayerNorm_9_1	LayerNorm	1 × 262144 × 7168
39	FFN_9	Feed Forward	1 × 262144 × 7168
40	Attention_10	Multi-Head Attention	1 × 262144 × 7168
41	Add_10	Add	1 × 262144 × 7168
42	LayerNorm_10_1	LayerNorm	1 × 262144 × 7168
43	FFN_10	Feed Forward	1 × 262144 × 7168
44	Attention_11	Multi-Head Attention	1 × 262144 × 7168
45	Add_11	Add	1 × 262144 × 7168
46	LayerNorm_11_1	LayerNorm	1 × 262144 × 7168
47	FFN_11	Feed Forward	1 × 262144 × 7168
48	Attention_12	Multi-Head Attention	1 × 262144 × 7168
49	Add_12	Add	1 × 262144 × 7168
50	LayerNorm_12_1	LayerNorm	1 × 262144 × 7168
51	FFN_12	Feed Forward	1 × 262144 × 7168
52	Attention_13	Multi-Head Attention	1 × 262144 × 7168
53	Add_13	Add	1 × 262144 × 7168
54	LayerNorm_13_1	LayerNorm	1 × 262144 × 7168
55	FFN_13	Feed Forward	1 × 262144 × 7168
56	Attention_14	Multi-Head Attention	1 × 262144 × 7168
57	Add_14	Add	1 × 262144 × 7168
58	LayerNorm_14_1	LayerNorm	1 × 262144 × 7168
59	FFN_14	Feed Forward	1 × 262144 × 7168
60	Attention_15	Multi-Head Attention	1 × 262144 × 7168
61	Add_15	Add	1 × 262144 × 7168
62	LayerNorm_15_1	LayerNorm	1 × 262144 × 7168
63	FFN_15	Feed Forward	1 × 262144 × 7168
64	Attention_16	Multi-Head Attention	1 × 262144 × 7168
65	Add_16	Add	1 × 262144 × 7168
66	LayerNorm_16_1	LayerNorm	1 × 262144 × 7168
67	FFN_16	Feed Forward	1 × 262144 × 7168
68	Attention_17	Multi-Head Attention	1 × 262144 × 7168
69	Add_17	Add	1 × 262144 × 7168
70	LayerNorm_17_1	LayerNorm	1 × 262144 × 7168
71	FFN_17	Feed Forward	1 × 262144 × 7168
72	Attention_18	Multi-Head Attention	1 × 262144 × 7168
73	Add_18	Add	1 × 262144 × 7168
74	LayerNorm_18_1	LayerNorm	1 × 262144 × 7168
75	FFN_18	Feed Forward	1 × 262144 × 7168
76	Attention_19	Multi-Head Attention	1 × 262144 × 7168
77	Add_19	Add	1 × 262144 × 7168
78	LayerNorm_19_1	LayerNorm	1 × 262144 × 7168
79	FFN_19	Feed Forward	1 × 262144 × 7168
80	Attention_20	Multi-Head Attention	1 × 262144 × 7168
81	Add_20	Add	1 × 262144 × 7168
82	LayerNorm_20_1	LayerNorm	1 × 262144 × 7168
83	FFN_20	Feed Forward	1 × 262144 × 7168
84	Attention_21	Multi-Head Attention	1 × 262144 × 7168
85	Add_21	Add	1 × 262144 × 7168
86	LayerNorm_21_1	LayerNorm	1 × 262144 × 7168
87	FFN_21	Feed Forward	1 × 262144 × 7168
88	Attention_22	Multi-Head Attention	1 × 262144 × 7168
89	Add_22	Add	1 × 262144 × 7168
90	LayerNorm_22_1	LayerNorm	1 × 262144 × 7168
91	FFN_22	Feed Forward	1 × 262144 × 7168
92	Attention_23	Multi-Head Attention	1 × 262144 × 7168
93	Add_23	Add	1 × 262144 × 7168
94	LayerNorm_23_1	LayerNorm	1 × 262144 × 7168
95	FFN_23	Feed Forward	1 × 262144 × 7168
96	Attention_24	Multi-Head Attention	1 × 262144 × 7168
97	Add_24	Add	1 × 262144 × 7168
98	LayerNorm_24_1	LayerNorm	1 × 262144 × 7168
99	FFN_24	Feed Forward	1 × 262144 × 7168
100	Attention_25	Multi-Head Attention	1 × 262144 × 7168
101	Add_25	Add	1 × 262144 × 7168
102	LayerNorm_25_1	LayerNorm	1 × 262144 × 7168
103	FFN_25	Feed Forward	1 × 262144 × 7168
104	Attention_26	Multi-Head Attention	1 × 262144 × 7168
105	Add_26	Add	1 × 262144 × 7168
106	LayerNorm_26_1	LayerNorm	1 × 262144 × 7168
107	FFN_26	Feed Forward	1 × 262144 × 7168
108	Attention_27	Multi-Head Attention	1 × 262144 × 7168
109	Add_27	Add	1 × 262144 × 7168
110	LayerNorm_27_1	LayerNorm	1 × 262144 × 7168
111	FFN_27	Feed Forward	1 × 262144 × 7168
112	Attention_28	Multi-Head Attention	1 × 262144 × 7168
113	Add_28	Add	1 × 262144 × 7168
114	LayerNorm_28_1	LayerNorm	1 × 262144 × 7168
115	FFN_28	Feed Forward	1 × 262144 × 7168
116	Attention_29	Multi-Head Attention	1 × 262144 × 7168
117	Add_29	Add	1 × 262144 × 7168
118	LayerNorm_29_1	LayerNorm	1 × 262144 × 7168
119	FFN_29	Feed Forward	1 × 262144 × 7168
120	Attention_30	Multi-Head Attention	1 × 262144 × 7168
121	Add_30	Add	1 × 262144 × 7168
122	LayerNorm_30_1	LayerNorm	1 × 262144 × 7168
123	FFN_30	Feed Forward	1 × 262144 × 7168
124	Attention_31	Multi-Head Attention	1 × 262144 × 7168
125	Add_31	Add	1 × 262144 × 7168
126	LayerNorm_31_1	LayerNorm	1 × 262144 × 7168
127	FFN_31	Feed Forward	1 × 262144 × 7168
128	Attention_32	Multi-Head Attention	1 × 262144 × 7168
129	Add_32	Add	1 × 262144 × 7168
130	LayerNorm_32_1	LayerNorm	1 × 262144 × 7168
131	FFN_32	Feed Forward	1 × 262144 × 7168
132	Attention_33	Multi-Head Attention	1 × 262144 × 7168
133	Add_33	Add	1 × 262144 × 7168
134	LayerNorm_33_1	LayerNorm	1 × 262144 × 7168
135	FFN_33	Feed Forward	1 × 262144 × 7168
136	Attention_34	Multi-Head Attention	1 × 262144 × 7168
137	Add_34	Add	1 × 262144 × 7168
138	LayerNorm_34_1	LayerNorm	1 × 262144 × 7168
139	FFN_34	Feed Forward	1 × 262144 × 7168
140	Attention_35	Multi-Head Attention	1 × 262144 × 7168
141	Add_35	Add	1 × 262144 × 7168
142	LayerNorm_35_1	LayerNorm	1 × 262144 × 7168
143	FFN_35	Feed Forward	1 × 262144 × 7168
144	Attention_36	Multi-Head Attention	1 × 262144 × 7168
145	Add_36	Add	1 × 262144 × 7168
146	LayerNorm_36_1	LayerNorm	1 × 262144 × 7168
147	FFN_36	Feed Forward	1 × 262144 × 7168
148	Attention_37	Multi-Head Attention	1 × 262144 × 7168
149	Add_37	Add	1 × 262144 × 7168
150	LayerNorm_37_1	LayerNorm	1 × 262144 × 7168
151	FFN_37	Feed Forward	1 × 262144 × 7168
152	Attention_38	Multi-Head Attention	1 × 262144 × 7168
153	Add_38	Add	1 × 262144 × 7168
154	LayerNorm_38_1	LayerNorm	1 × 262144 × 7168
155	FFN_38	Feed Forward	1 × 262144 × 7168
156	Attention_39	Multi-Head Attention	1 × 262144 × 7168
157	Add_39	Add	1 × 262144 × 7168
158	LayerNorm_39_1	LayerNorm	1 × 262144 × 7168
159	FFN_39	Feed Forward	1 × 262144 × 7168
160	Attention_40	Multi-Head Attention	1 × 262144 × 7168
161	Add_40	Add	1 × 262144 × 7168
162	LayerNorm_40_1	LayerNorm	1 × 262144 × 7168
163	FFN_40	Feed Forward	1 × 262144 × 7168
164	Attention_41	Multi-Head Attention	1 × 262144 × 7168
165	Add_41	Add	1 × 262144 × 7168
166	LayerNorm_41_1	LayerNorm	1 × 262144 × 7168
167	FFN_41	Feed Forward	1 × 262144 × 7168
168	Attention_42	Multi-Head Attention	1 × 262144 × 7168
169	Add_42	Add	1 × 262144 × 7168
170	LayerNorm_42_1	LayerNorm	1 × 262144 × 7168
171	FFN_42	Feed Forward	1 × 262144 × 7168
172	Attention_43	Multi-Head Attention	1 × 262144 × 7168
173	Add_43	Add	1 × 262144 × 7168
174	LayerNorm_43_1	LayerNorm	1 × 262144 × 7168
175	FFN_43	Feed Forward	1 × 262144 × 7168
176	Attention_44	Multi-Head Attention	1 × 262144 × 7168
177	Add_44	Add	1 × 262144 × 7168
178	LayerNorm_44_1	LayerNorm	1 × 262144 × 7168
179	FFN_44	Feed Forward	1 × 262144 × 7168
180	Attention_45	Multi-Head Attention	1 × 262144 × 7168
181	Add_45	Add	1 × 262144 × 7168
182	LayerNorm_45_1	LayerNorm	1 × 262144 × 7168
183	FFN_45	Feed Forward	1 × 262144 × 7168
184	Attention_46	Multi-Head Attention	1 × 262144 × 7168
185	Add_46	Add	1 × 262144 × 7168
186	LayerNorm_46_1	LayerNorm	1 × 262144 × 7168
187	FFN_46	Feed Forward	1 × 262144 × 7168
188	Attention_47	Multi-Head Attention	1 × 262144 × 7168
189	Add_47	Add	1 × 262144 × 7168
190	LayerNorm_47_1	LayerNorm	1 × 262144 × 7168
191	FFN_47	Feed Forward	1 × 262144 × 7168
192	Attention_48	Multi-Head Attention	1 × 262144 × 7168
193	Add_48	Add	1 × 262144 × 7168
194	LayerNorm_48_1	LayerNorm	1 × 262144 × 7168
195	FFN_48	Feed Forward	1 × 262144 × 7168
196	Attention_49	Multi-Head Attention	1 × 262144 × 7168
197	Add_49	Add	1 × 262144 × 7168
198	LayerNorm_49_1	LayerNorm	1 × 262144 × 7168
199	FFN_49	Feed Forward	1 × 262144 × 7168
200	Attention_50	Multi-Head Attention	1 × 262144 × 7168
201	Add_50	Add	1 × 262144 × 7168
202	LayerNorm_50_1	LayerNorm	1 × 262144 × 7168
203	FFN_50	Feed Forward	1 × 262144 × 7168
204	Attention_51	Multi-Head Attention	1 × 262144 × 7168
205	Add_51	Add	1 × 262144 × 7168
206	LayerNorm_51_1	LayerNorm	1 × 262144 × 7168
207	FFN_51	Feed Forward	1 × 262144 × 7168
208	Attention_52	Multi-Head Attention	1 × 262144 × 7168
209	Add_52	Add	1 × 262144 × 7168
210	LayerNorm_52_1	LayerNorm	1 × 262144 × 7168
211	FFN_52	Feed Forward	1 × 262144 × 7168
212	Attention_53	Multi-Head Attention	1 × 262144 × 7168
213	Add_53	Add	1 × 262144 × 7168
214	LayerNorm_53_1	LayerNorm	1 × 262144 × 7168
215	FFN_53	Feed Forward	1 × 262144 × 7168
216	Attention_54	Multi-Head Attention	1 × 262144 × 7168
217	Add_54	Add	1 × 262144 × 7168
218	LayerNorm_54_1	LayerNorm	1 × 262144 × 7168
219	FFN_54	Feed Forward	1 × 262144 × 7168
220	Attention_55	Multi-Head Attention	1 × 262144 × 7168
221	Add_55	Add	1 × 262144 × 7168
222	LayerNorm_55_1	LayerNorm	1 × 262144 × 7168
223	FFN_55	Feed Forward	1 × 262144 × 7168
224	Attention_56	Multi-Head Attention	1 × 262144 × 7168
225	Add_56	Add	1 × 262144 × 7168
226	LayerNorm_56_1	LayerNorm	1 × 262144 × 7168
227	FFN_56	Feed Forward	1 × 262144 × 7168
228	Attention_57	Multi-Head Attention	1 × 262144 × 7168
229	Add_57	Add	1 × 262144 × 7168
230	LayerNorm_57_1	LayerNorm	1 × 262144 × 7168
231	FFN_57	Feed Forward	1 × 262144 × 7168
232	Attention_58	Multi-Head Attention	1 × 262144 × 7168
233	Add_58	Add	1 × 262144 × 7168
234	LayerNorm_58_1	LayerNorm	1 × 262144 × 7168
235	FFN_58	Feed Forward	1 × 262144 × 7168
236	Attention_59	Multi-Head Attention	1 × 262144 × 7168
237	Add_59	Add	1 × 262144 × 7168
238	LayerNorm_59_1	LayerNorm	1 × 262144 × 7168
239	FFN_59	Feed Forward	1 × 262144 × 7168
240	Attention_60	Multi-Head Attention	1 × 262144 × 7168
241	Add_60	Add	1 × 262144 × 7168
242	LayerNorm_60_1	LayerNorm	1 × 262144 × 7168
243	FFN_60	Feed Forward	1 × 262144 × 7168
244	Attention_61	Multi-Head Attention	1 × 262144 × 7168
245	Add_61	Add	1 × 262144 × 7168
246	LayerNorm_61_1	LayerNorm	1 × 262144 × 7168
247	FFN_61	Feed Forward	1 × 262144 × 7168
248	Output	Output	1 × 262144 × 7168

What the verifier says

info61 attention layers at embedDim 7168 cache full per-head K/V: about 1708 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 61 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 61 attention layers this design caches 1708 KB per token, so a single 8,192-token sequence needs ~14.3 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 moonshotai/Kimi-K2.6 --plan --share

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

Other kimi_k25 checkpoints

Kimi-K2.5

31.71B derived · -96.9% against the checkpoint

Kimi-K2.7-Code

31.71B derived · -96.9% against the checkpoint