Models / moss_transcribe_diarize

MOSS-Transcribe-Diarize

Reconstructed from its own config.json with no weights read. 251K 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

899M

899,434,496 parameters

In the published checkpoint

909M

908,513,280 scalars · safetensors.total, read 2026-09-02

Delta

-1.00%

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

270

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$39182.39

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

Structure

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

	Layer	Type	Output shape
1	Input	Input	1 × 131072
2	Embedding	Embedding	1 × 131072 × 1024
3	RoPE	RoPE	1 × 131072 × 1024
4	Audio input	Input	1 × 80 × 1
5	Audio feature embed	Projection	1 × 80 × 1024
6	Audio_LN_1	LayerNorm	1 × 80 × 1024
7	Audio_Attn_1	Multi-Head Attention	1 × 80 × 1024
8	Audio_Add_1	Add	1 × 80 × 1024
9	Audio_FFN_1	Feed Forward	1 × 80 × 1024
10	Audio_LN_2	LayerNorm	1 × 80 × 1024
11	Audio_Attn_2	Multi-Head Attention	1 × 80 × 1024
12	Audio_Add_2	Add	1 × 80 × 1024
13	Audio_FFN_2	Feed Forward	1 × 80 × 1024
14	Audio_LN_3	LayerNorm	1 × 80 × 1024
15	Audio_Attn_3	Multi-Head Attention	1 × 80 × 1024
16	Audio_Add_3	Add	1 × 80 × 1024
17	Audio_FFN_3	Feed Forward	1 × 80 × 1024
18	Audio_LN_4	LayerNorm	1 × 80 × 1024
19	Audio_Attn_4	Multi-Head Attention	1 × 80 × 1024
20	Audio_Add_4	Add	1 × 80 × 1024
21	Audio_FFN_4	Feed Forward	1 × 80 × 1024
22	Audio_LN_5	LayerNorm	1 × 80 × 1024
23	Audio_Attn_5	Multi-Head Attention	1 × 80 × 1024
24	Audio_Add_5	Add	1 × 80 × 1024
25	Audio_FFN_5	Feed Forward	1 × 80 × 1024
26	Audio_LN_6	LayerNorm	1 × 80 × 1024
27	Audio_Attn_6	Multi-Head Attention	1 × 80 × 1024
28	Audio_Add_6	Add	1 × 80 × 1024
29	Audio_FFN_6	Feed Forward	1 × 80 × 1024
30	Audio_LN_7	LayerNorm	1 × 80 × 1024
31	Audio_Attn_7	Multi-Head Attention	1 × 80 × 1024
32	Audio_Add_7	Add	1 × 80 × 1024
33	Audio_FFN_7	Feed Forward	1 × 80 × 1024
34	Audio_LN_8	LayerNorm	1 × 80 × 1024
35	Audio_Attn_8	Multi-Head Attention	1 × 80 × 1024
36	Audio_Add_8	Add	1 × 80 × 1024
37	Audio_FFN_8	Feed Forward	1 × 80 × 1024
38	Audio_LN_9	LayerNorm	1 × 80 × 1024
39	Audio_Attn_9	Multi-Head Attention	1 × 80 × 1024
40	Audio_Add_9	Add	1 × 80 × 1024
41	Audio_FFN_9	Feed Forward	1 × 80 × 1024
42	Audio_LN_10	LayerNorm	1 × 80 × 1024
43	Audio_Attn_10	Multi-Head Attention	1 × 80 × 1024
44	Audio_Add_10	Add	1 × 80 × 1024
45	Audio_FFN_10	Feed Forward	1 × 80 × 1024
46	Audio_LN_11	LayerNorm	1 × 80 × 1024
47	Audio_Attn_11	Multi-Head Attention	1 × 80 × 1024
48	Audio_Add_11	Add	1 × 80 × 1024
49	Audio_FFN_11	Feed Forward	1 × 80 × 1024
50	Audio_LN_12	LayerNorm	1 × 80 × 1024
51	Audio_Attn_12	Multi-Head Attention	1 × 80 × 1024
52	Audio_Add_12	Add	1 × 80 × 1024
53	Audio_FFN_12	Feed Forward	1 × 80 × 1024
54	Audio_LN_13	LayerNorm	1 × 80 × 1024
55	Audio_Attn_13	Multi-Head Attention	1 × 80 × 1024
56	Audio_Add_13	Add	1 × 80 × 1024
57	Audio_FFN_13	Feed Forward	1 × 80 × 1024
58	Audio_LN_14	LayerNorm	1 × 80 × 1024
59	Audio_Attn_14	Multi-Head Attention	1 × 80 × 1024
60	Audio_Add_14	Add	1 × 80 × 1024
61	Audio_FFN_14	Feed Forward	1 × 80 × 1024
62	Audio_LN_15	LayerNorm	1 × 80 × 1024
63	Audio_Attn_15	Multi-Head Attention	1 × 80 × 1024
64	Audio_Add_15	Add	1 × 80 × 1024
65	Audio_FFN_15	Feed Forward	1 × 80 × 1024
66	Audio_LN_16	LayerNorm	1 × 80 × 1024
67	Audio_Attn_16	Multi-Head Attention	1 × 80 × 1024
68	Audio_Add_16	Add	1 × 80 × 1024
69	Audio_FFN_16	Feed Forward	1 × 80 × 1024
70	Audio_LN_17	LayerNorm	1 × 80 × 1024
71	Audio_Attn_17	Multi-Head Attention	1 × 80 × 1024
72	Audio_Add_17	Add	1 × 80 × 1024
73	Audio_FFN_17	Feed Forward	1 × 80 × 1024
74	Audio_LN_18	LayerNorm	1 × 80 × 1024
75	Audio_Attn_18	Multi-Head Attention	1 × 80 × 1024
76	Audio_Add_18	Add	1 × 80 × 1024
77	Audio_FFN_18	Feed Forward	1 × 80 × 1024
78	Audio_LN_19	LayerNorm	1 × 80 × 1024
79	Audio_Attn_19	Multi-Head Attention	1 × 80 × 1024
80	Audio_Add_19	Add	1 × 80 × 1024
81	Audio_FFN_19	Feed Forward	1 × 80 × 1024
82	Audio_LN_20	LayerNorm	1 × 80 × 1024
83	Audio_Attn_20	Multi-Head Attention	1 × 80 × 1024
84	Audio_Add_20	Add	1 × 80 × 1024
85	Audio_FFN_20	Feed Forward	1 × 80 × 1024
86	Audio_LN_21	LayerNorm	1 × 80 × 1024
87	Audio_Attn_21	Multi-Head Attention	1 × 80 × 1024
88	Audio_Add_21	Add	1 × 80 × 1024
89	Audio_FFN_21	Feed Forward	1 × 80 × 1024
90	Audio_LN_22	LayerNorm	1 × 80 × 1024
91	Audio_Attn_22	Multi-Head Attention	1 × 80 × 1024
92	Audio_Add_22	Add	1 × 80 × 1024
93	Audio_FFN_22	Feed Forward	1 × 80 × 1024
94	Audio_LN_23	LayerNorm	1 × 80 × 1024
95	Audio_Attn_23	Multi-Head Attention	1 × 80 × 1024
96	Audio_Add_23	Add	1 × 80 × 1024
97	Audio_FFN_23	Feed Forward	1 × 80 × 1024
98	Audio_LN_24	LayerNorm	1 × 80 × 1024
99	Audio_Attn_24	Multi-Head Attention	1 × 80 × 1024
100	Audio_Add_24	Add	1 × 80 × 1024
101	Audio_FFN_24	Feed Forward	1 × 80 × 1024
102	Audio projector	Projection	1 × 80 × 1024
103	Audio tokens	Reshape	1 × 256 × 1024
104	Multimodal fusion (concat tokens)	Concatenate	1 × 131328 × 1024
105	RMSNorm_1_1	RMSNorm	1 × 131328 × 1024
106	Attention_1	Grouped Query Attn	1 × 131328 × 1024
107	Add_1_attn	Add	1 × 131328 × 1024
108	RMSNorm_1_2	RMSNorm	1 × 131328 × 1024
109	FFN_1	SwiGLU	1 × 131328 × 1024
110	Add_1_ffn	Add	1 × 131328 × 1024
111	RMSNorm_2_1	RMSNorm	1 × 131328 × 1024
112	Attention_2	Grouped Query Attn	1 × 131328 × 1024
113	Add_2_attn	Add	1 × 131328 × 1024
114	RMSNorm_2_2	RMSNorm	1 × 131328 × 1024
115	FFN_2	SwiGLU	1 × 131328 × 1024
116	Add_2_ffn	Add	1 × 131328 × 1024
117	RMSNorm_3_1	RMSNorm	1 × 131328 × 1024
118	Attention_3	Grouped Query Attn	1 × 131328 × 1024
119	Add_3_attn	Add	1 × 131328 × 1024
120	RMSNorm_3_2	RMSNorm	1 × 131328 × 1024
121	FFN_3	SwiGLU	1 × 131328 × 1024
122	Add_3_ffn	Add	1 × 131328 × 1024
123	RMSNorm_4_1	RMSNorm	1 × 131328 × 1024
124	Attention_4	Grouped Query Attn	1 × 131328 × 1024
125	Add_4_attn	Add	1 × 131328 × 1024
126	RMSNorm_4_2	RMSNorm	1 × 131328 × 1024
127	FFN_4	SwiGLU	1 × 131328 × 1024
128	Add_4_ffn	Add	1 × 131328 × 1024
129	RMSNorm_5_1	RMSNorm	1 × 131328 × 1024
130	Attention_5	Grouped Query Attn	1 × 131328 × 1024
131	Add_5_attn	Add	1 × 131328 × 1024
132	RMSNorm_5_2	RMSNorm	1 × 131328 × 1024
133	FFN_5	SwiGLU	1 × 131328 × 1024
134	Add_5_ffn	Add	1 × 131328 × 1024
135	RMSNorm_6_1	RMSNorm	1 × 131328 × 1024
136	Attention_6	Grouped Query Attn	1 × 131328 × 1024
137	Add_6_attn	Add	1 × 131328 × 1024
138	RMSNorm_6_2	RMSNorm	1 × 131328 × 1024
139	FFN_6	SwiGLU	1 × 131328 × 1024
140	Add_6_ffn	Add	1 × 131328 × 1024
141	RMSNorm_7_1	RMSNorm	1 × 131328 × 1024
142	Attention_7	Grouped Query Attn	1 × 131328 × 1024
143	Add_7_attn	Add	1 × 131328 × 1024
144	RMSNorm_7_2	RMSNorm	1 × 131328 × 1024
145	FFN_7	SwiGLU	1 × 131328 × 1024
146	Add_7_ffn	Add	1 × 131328 × 1024
147	RMSNorm_8_1	RMSNorm	1 × 131328 × 1024
148	Attention_8	Grouped Query Attn	1 × 131328 × 1024
149	Add_8_attn	Add	1 × 131328 × 1024
150	RMSNorm_8_2	RMSNorm	1 × 131328 × 1024
151	FFN_8	SwiGLU	1 × 131328 × 1024
152	Add_8_ffn	Add	1 × 131328 × 1024
153	RMSNorm_9_1	RMSNorm	1 × 131328 × 1024
154	Attention_9	Grouped Query Attn	1 × 131328 × 1024
155	Add_9_attn	Add	1 × 131328 × 1024
156	RMSNorm_9_2	RMSNorm	1 × 131328 × 1024
157	FFN_9	SwiGLU	1 × 131328 × 1024
158	Add_9_ffn	Add	1 × 131328 × 1024
159	RMSNorm_10_1	RMSNorm	1 × 131328 × 1024
160	Attention_10	Grouped Query Attn	1 × 131328 × 1024
161	Add_10_attn	Add	1 × 131328 × 1024
162	RMSNorm_10_2	RMSNorm	1 × 131328 × 1024
163	FFN_10	SwiGLU	1 × 131328 × 1024
164	Add_10_ffn	Add	1 × 131328 × 1024
165	RMSNorm_11_1	RMSNorm	1 × 131328 × 1024
166	Attention_11	Grouped Query Attn	1 × 131328 × 1024
167	Add_11_attn	Add	1 × 131328 × 1024
168	RMSNorm_11_2	RMSNorm	1 × 131328 × 1024
169	FFN_11	SwiGLU	1 × 131328 × 1024
170	Add_11_ffn	Add	1 × 131328 × 1024
171	RMSNorm_12_1	RMSNorm	1 × 131328 × 1024
172	Attention_12	Grouped Query Attn	1 × 131328 × 1024
173	Add_12_attn	Add	1 × 131328 × 1024
174	RMSNorm_12_2	RMSNorm	1 × 131328 × 1024
175	FFN_12	SwiGLU	1 × 131328 × 1024
176	Add_12_ffn	Add	1 × 131328 × 1024
177	RMSNorm_13_1	RMSNorm	1 × 131328 × 1024
178	Attention_13	Grouped Query Attn	1 × 131328 × 1024
179	Add_13_attn	Add	1 × 131328 × 1024
180	RMSNorm_13_2	RMSNorm	1 × 131328 × 1024
181	FFN_13	SwiGLU	1 × 131328 × 1024
182	Add_13_ffn	Add	1 × 131328 × 1024
183	RMSNorm_14_1	RMSNorm	1 × 131328 × 1024
184	Attention_14	Grouped Query Attn	1 × 131328 × 1024
185	Add_14_attn	Add	1 × 131328 × 1024
186	RMSNorm_14_2	RMSNorm	1 × 131328 × 1024
187	FFN_14	SwiGLU	1 × 131328 × 1024
188	Add_14_ffn	Add	1 × 131328 × 1024
189	RMSNorm_15_1	RMSNorm	1 × 131328 × 1024
190	Attention_15	Grouped Query Attn	1 × 131328 × 1024
191	Add_15_attn	Add	1 × 131328 × 1024
192	RMSNorm_15_2	RMSNorm	1 × 131328 × 1024
193	FFN_15	SwiGLU	1 × 131328 × 1024
194	Add_15_ffn	Add	1 × 131328 × 1024
195	RMSNorm_16_1	RMSNorm	1 × 131328 × 1024
196	Attention_16	Grouped Query Attn	1 × 131328 × 1024
197	Add_16_attn	Add	1 × 131328 × 1024
198	RMSNorm_16_2	RMSNorm	1 × 131328 × 1024
199	FFN_16	SwiGLU	1 × 131328 × 1024
200	Add_16_ffn	Add	1 × 131328 × 1024
201	RMSNorm_17_1	RMSNorm	1 × 131328 × 1024
202	Attention_17	Grouped Query Attn	1 × 131328 × 1024
203	Add_17_attn	Add	1 × 131328 × 1024
204	RMSNorm_17_2	RMSNorm	1 × 131328 × 1024
205	FFN_17	SwiGLU	1 × 131328 × 1024
206	Add_17_ffn	Add	1 × 131328 × 1024
207	RMSNorm_18_1	RMSNorm	1 × 131328 × 1024
208	Attention_18	Grouped Query Attn	1 × 131328 × 1024
209	Add_18_attn	Add	1 × 131328 × 1024
210	RMSNorm_18_2	RMSNorm	1 × 131328 × 1024
211	FFN_18	SwiGLU	1 × 131328 × 1024
212	Add_18_ffn	Add	1 × 131328 × 1024
213	RMSNorm_19_1	RMSNorm	1 × 131328 × 1024
214	Attention_19	Grouped Query Attn	1 × 131328 × 1024
215	Add_19_attn	Add	1 × 131328 × 1024
216	RMSNorm_19_2	RMSNorm	1 × 131328 × 1024
217	FFN_19	SwiGLU	1 × 131328 × 1024
218	Add_19_ffn	Add	1 × 131328 × 1024
219	RMSNorm_20_1	RMSNorm	1 × 131328 × 1024
220	Attention_20	Grouped Query Attn	1 × 131328 × 1024
221	Add_20_attn	Add	1 × 131328 × 1024
222	RMSNorm_20_2	RMSNorm	1 × 131328 × 1024
223	FFN_20	SwiGLU	1 × 131328 × 1024
224	Add_20_ffn	Add	1 × 131328 × 1024
225	RMSNorm_21_1	RMSNorm	1 × 131328 × 1024
226	Attention_21	Grouped Query Attn	1 × 131328 × 1024
227	Add_21_attn	Add	1 × 131328 × 1024
228	RMSNorm_21_2	RMSNorm	1 × 131328 × 1024
229	FFN_21	SwiGLU	1 × 131328 × 1024
230	Add_21_ffn	Add	1 × 131328 × 1024
231	RMSNorm_22_1	RMSNorm	1 × 131328 × 1024
232	Attention_22	Grouped Query Attn	1 × 131328 × 1024
233	Add_22_attn	Add	1 × 131328 × 1024
234	RMSNorm_22_2	RMSNorm	1 × 131328 × 1024
235	FFN_22	SwiGLU	1 × 131328 × 1024
236	Add_22_ffn	Add	1 × 131328 × 1024
237	RMSNorm_23_1	RMSNorm	1 × 131328 × 1024
238	Attention_23	Grouped Query Attn	1 × 131328 × 1024
239	Add_23_attn	Add	1 × 131328 × 1024
240	RMSNorm_23_2	RMSNorm	1 × 131328 × 1024
241	FFN_23	SwiGLU	1 × 131328 × 1024
242	Add_23_ffn	Add	1 × 131328 × 1024
243	RMSNorm_24_1	RMSNorm	1 × 131328 × 1024
244	Attention_24	Grouped Query Attn	1 × 131328 × 1024
245	Add_24_attn	Add	1 × 131328 × 1024
246	RMSNorm_24_2	RMSNorm	1 × 131328 × 1024
247	FFN_24	SwiGLU	1 × 131328 × 1024
248	Add_24_ffn	Add	1 × 131328 × 1024
249	RMSNorm_25_1	RMSNorm	1 × 131328 × 1024
250	Attention_25	Grouped Query Attn	1 × 131328 × 1024
251	Add_25_attn	Add	1 × 131328 × 1024
252	RMSNorm_25_2	RMSNorm	1 × 131328 × 1024
253	FFN_25	SwiGLU	1 × 131328 × 1024
254	Add_25_ffn	Add	1 × 131328 × 1024
255	RMSNorm_26_1	RMSNorm	1 × 131328 × 1024
256	Attention_26	Grouped Query Attn	1 × 131328 × 1024
257	Add_26_attn	Add	1 × 131328 × 1024
258	RMSNorm_26_2	RMSNorm	1 × 131328 × 1024
259	FFN_26	SwiGLU	1 × 131328 × 1024
260	Add_26_ffn	Add	1 × 131328 × 1024
261	RMSNorm_27_1	RMSNorm	1 × 131328 × 1024
262	Attention_27	Grouped Query Attn	1 × 131328 × 1024
263	Add_27_attn	Add	1 × 131328 × 1024
264	RMSNorm_27_2	RMSNorm	1 × 131328 × 1024
265	FFN_27	SwiGLU	1 × 131328 × 1024
266	Add_27_ffn	Add	1 × 131328 × 1024
267	RMSNorm_28_1	RMSNorm	1 × 131328 × 1024
268	Attention_28	Grouped Query Attn	1 × 131328 × 1024
269	Add_28_attn	Add	1 × 131328 × 1024
270	RMSNorm_28_2	RMSNorm	1 × 131328 × 1024
271	FFN_28	SwiGLU	1 × 131328 × 1024
272	Add_28_ffn	Add	1 × 131328 × 1024
273	Output	Output	1 × 131328 × 1024

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 52 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 OpenMOSS-Team/MOSS-Transcribe-Diarize --plan --share

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