Qwen3.6-35B-A3B-DFlash

Reconstructed from its own config.json with no weights read. 208K 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

1.37B

1,369,715,200 parameters

In the published checkpoint

386M

385,906,176 scalars · safetensors.total, read 2026-06-19

Delta

+255%

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

Will it forward-pass

Yes

Priced on

A10G (24GB)

Est. one run

$71655.72

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

Structure

42 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 × 2048
3	RoPE	RoPE	1 × 262144 × 2048
4	RMSNorm_1_1	RMSNorm	1 × 262144 × 2048
5	Attention_1	Grouped Query Attn	1 × 262144 × 2048
6	Add_1_attn	Add	1 × 262144 × 2048
7	RMSNorm_1_2	RMSNorm	1 × 262144 × 2048
8	FFN_1	SwiGLU	1 × 262144 × 2048
9	Add_1_ffn	Add	1 × 262144 × 2048
10	RMSNorm_2_1	RMSNorm	1 × 262144 × 2048
11	Attention_2	Grouped Query Attn	1 × 262144 × 2048
12	Add_2_attn	Add	1 × 262144 × 2048
13	RMSNorm_2_2	RMSNorm	1 × 262144 × 2048
14	FFN_2	SwiGLU	1 × 262144 × 2048
15	Add_2_ffn	Add	1 × 262144 × 2048
16	RMSNorm_3_1	RMSNorm	1 × 262144 × 2048
17	Attention_3	Grouped Query Attn	1 × 262144 × 2048
18	Add_3_attn	Add	1 × 262144 × 2048
19	RMSNorm_3_2	RMSNorm	1 × 262144 × 2048
20	FFN_3	SwiGLU	1 × 262144 × 2048
21	Add_3_ffn	Add	1 × 262144 × 2048
22	RMSNorm_4_1	RMSNorm	1 × 262144 × 2048
23	Attention_4	Grouped Query Attn	1 × 262144 × 2048
24	Add_4_attn	Add	1 × 262144 × 2048
25	RMSNorm_4_2	RMSNorm	1 × 262144 × 2048
26	FFN_4	SwiGLU	1 × 262144 × 2048
27	Add_4_ffn	Add	1 × 262144 × 2048
28	RMSNorm_5_1	RMSNorm	1 × 262144 × 2048
29	Attention_5	Grouped Query Attn	1 × 262144 × 2048
30	Add_5_attn	Add	1 × 262144 × 2048
31	RMSNorm_5_2	RMSNorm	1 × 262144 × 2048
32	FFN_5	SwiGLU	1 × 262144 × 2048
33	Add_5_ffn	Add	1 × 262144 × 2048
34	RMSNorm_6_1	RMSNorm	1 × 262144 × 2048
35	Attention_6	Grouped Query Attn	1 × 262144 × 2048
36	Add_6_attn	Add	1 × 262144 × 2048
37	RMSNorm_6_2	RMSNorm	1 × 262144 × 2048
38	FFN_6	SwiGLU	1 × 262144 × 2048
39	Add_6_ffn	Add	1 × 262144 × 2048
40	Final_RMSNorm	RMSNorm	1 × 262144 × 2048
41	LM_Head	Linear	1 × 262144 × 248320
42	Output	Output	1 × 262144 × 248320

What the verifier says

No finding on the reconstructed graph. See the checks.

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 z-lab/Qwen3.6-35B-A3B-DFlash --plan --share

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

Other qwen3 checkpoints

jina-reranker-v3

596M derived · -0.13% against the checkpoint

Kimi-K3-DSpark

4.26B derived · +89.3% against the checkpoint

Qwen3-0.6B

596M derived · -20.7% against the checkpoint

Qwen3-1.7B

1.72B derived · -15.3% against the checkpoint