Datasets:

neurarch-ai
/

arch-verifier-grounding-264

taskId stringlengths 8 12	variant stringclasses 4 values	verifierScore int64 10 100	verifierBlocked bool 2 classes	constructs bool 2 classes	forward_ok bool 2 classes	trains bool 2 classes	initial_loss float64 0.7 7.55 ⌀	final_loss float64 0 3.92 ⌀	error stringlengths 24 65 ⌀
gen-mlp-0	clean	88	false	true	true	true	3.6112	0	null
gen-mlp-0	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x249 and 498x395)
gen-mlp-0	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-ae-1	clean	63	false	true	true	true	7.5508	0	null
gen-ae-1	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x817 and 1634x107)
gen-ae-1	disconnect	10	true	true	false	false	null	null	node enc2 has no computed input (disconnected)
gen-cnn-2	clean	80	false	true	true	false	4.6266	3.9201	null
gen-cnn-2	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x41 and 82x100)
gen-cnn-2	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-txf-3	clean	60	false	true	true	true	3.9606	0.013	null
gen-txf-3	divisibility	15	true	false	false	false	null	null	embed_dim must be divisible by num_heads
gen-txf-3	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x164 and 328x52)
gen-txf-3	disconnect	10	true	true	false	false	null	null	node attn1 has no computed input (disconnected)
gen-gqa-4	clean	81	false	true	true	true	3.3635	0.3238	null
gen-gqa-4	divisibility	10	true	false	false	false	null	null	embedDim % numHeads != 0
gen-gqa-4	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x48 and 96x27)
gen-gqa-4	disconnect	10	true	true	false	false	null	null	node attn1 has no computed input (disconnected)
gen-fix-5	clean	60	false	true	true	true	4.2782	0.0001	null
gen-fix-5	divisibility	15	true	false	false	false	null	null	numHeads % numKVHeads != 0
gen-fix-5	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x336 and 672x72)
gen-fix-5	disconnect	10	true	true	false	false	null	null	node attn has no computed input (disconnected)
gen-trim-6	clean	80	false	true	true	true	3.3614	0.0011	null
gen-trim-6	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x252 and 504x252)
gen-trim-6	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-norm-7	clean	100	false	true	true	true	3.3353	0.0026	null
gen-norm-7	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x253 and 506x437)
gen-norm-7	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-tower-8	clean	88	false	true	true	true	0.9051	0.0008	null
gen-tower-8	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x67 and 134x67)
gen-tower-8	disconnect	15	true	true	false	false	null	null	node user_act has no computed input (disconnected)
gen-grow-9	clean	80	false	true	true	true	3.4335	0	null
gen-grow-9	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x1061 and 2122x1061)
gen-grow-9	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-mlp-10	clean	84	false	true	true	true	5.0861	0	null
gen-mlp-10	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x917 and 1834x185)
gen-mlp-10	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-ae-11	clean	63	false	true	true	true	7.483	0	null
gen-ae-11	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x350 and 700x138)
gen-ae-11	disconnect	10	true	true	false	false	null	null	node enc2 has no computed input (disconnected)
gen-cnn-12	clean	80	false	true	true	false	2.9339	2.7996	null
gen-cnn-12	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x50 and 100x19)
gen-cnn-12	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-txf-13	clean	60	false	true	true	true	4.5138	2.0971	null
gen-txf-13	divisibility	15	true	false	false	false	null	null	embed_dim must be divisible by num_heads
gen-txf-13	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x88 and 176x90)
gen-txf-13	disconnect	10	true	true	false	false	null	null	node attn1 has no computed input (disconnected)
gen-gqa-14	clean	81	false	true	true	true	4.2764	0.0012	null
gen-gqa-14	divisibility	10	true	false	false	false	null	null	embedDim % numHeads != 0
gen-gqa-14	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x376 and 752x55)
gen-gqa-14	disconnect	10	true	true	false	false	null	null	node attn1 has no computed input (disconnected)
gen-fix-15	clean	60	false	true	true	true	4.3144	0	null
gen-fix-15	divisibility	15	true	false	false	false	null	null	embed_dim must be divisible by num_heads
gen-fix-15	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x380 and 760x75)
gen-fix-15	disconnect	10	true	true	false	false	null	null	node attn has no computed input (disconnected)
gen-trim-16	clean	80	false	true	true	true	3.2679	0.0006	null
gen-trim-16	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x315 and 630x315)
gen-trim-16	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-norm-17	clean	100	false	true	true	true	3.2302	0.0344	null
gen-norm-17	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x159 and 318x329)
gen-norm-17	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-tower-18	clean	88	false	true	true	true	0.9806	0.0008	null
gen-tower-18	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x71 and 142x71)
gen-tower-18	disconnect	15	true	true	false	false	null	null	node user_act has no computed input (disconnected)
gen-grow-19	clean	80	false	true	true	true	3.3367	0.0012	null
gen-grow-19	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x674 and 1348x674)
gen-grow-19	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-mlp-20	clean	90	false	true	true	true	3.4968	0	null
gen-mlp-20	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x183 and 366x747)
gen-mlp-20	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-ae-21	clean	63	false	true	true	true	6.7042	0	null
gen-ae-21	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x341 and 682x242)
gen-ae-21	disconnect	10	true	true	false	false	null	null	node enc2 has no computed input (disconnected)
gen-cnn-22	clean	80	false	true	true	false	3.321	3.0106	null
gen-cnn-22	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x32 and 64x27)
gen-cnn-22	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-txf-23	clean	63	false	true	true	true	3.5299	0	null
gen-txf-23	divisibility	15	true	false	false	false	null	null	embed_dim must be divisible by num_heads
gen-txf-23	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x236 and 472x34)
gen-txf-23	disconnect	10	true	true	false	false	null	null	node attn1 has no computed input (disconnected)
gen-gqa-24	clean	77	false	true	true	true	3.8732	0.1053	null
gen-gqa-24	divisibility	10	true	false	false	false	null	null	embedDim % numHeads != 0
gen-gqa-24	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x72 and 144x39)
gen-gqa-24	disconnect	10	true	true	false	false	null	null	node attn1 has no computed input (disconnected)
gen-fix-25	clean	60	false	true	true	true	3.5305	0.0002	null
gen-fix-25	divisibility	10	true	false	false	false	null	null	embedDim % numHeads != 0
gen-fix-25	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x328 and 656x34)
gen-fix-25	disconnect	10	true	true	false	false	null	null	node attn has no computed input (disconnected)
gen-trim-26	clean	80	false	true	true	true	1.6005	0.0748	null
gen-trim-26	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x70 and 140x70)
gen-trim-26	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-norm-27	clean	94	false	true	true	true	3.5766	0.0232	null
gen-norm-27	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x431 and 862x197)
gen-norm-27	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-tower-28	clean	88	false	true	true	true	1.0042	0.0003	null
gen-tower-28	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x82 and 164x82)
gen-tower-28	disconnect	15	true	true	false	false	null	null	node user_act has no computed input (disconnected)
gen-grow-29	clean	80	false	true	true	true	3.6379	0	null
gen-grow-29	linear-mismatch	15	true	true	false	false	null	null	mat1 and mat2 shapes cannot be multiplied (64x916 and 1832x916)
gen-grow-29	disconnect	10	true	true	false	false	null	null	node act1 has no computed input (disconnected)
gen-mlp-30	clean	88	false	true	true	true	3.8564	0	null

End of preview. Expand in Data Studio

Verifier grounding study (264 graphs)

Clean reference architectures plus systematically corrupted variants (broken attention head divisibility, linear width mismatches, severed connections), each built as a real PyTorch model and run on a GPU. Every row pairs the static verifier verdict with what actually happened at runtime: whether the module constructed, whether the forward pass survived, whether training made progress, and the initial and final loss. 264 graphs, two seeds, torch 2.8.

Size: 264 graphs
Licence: MIT License
Canonical page: https://neurarch.com/d/grounding-264.html
On the site: https://neurarch.com/caught-bugs.html

Fields

verifier score
verifier blocked
constructs
forward pass ok
makes training progress
initial loss
final loss

What this dataset is not

The corruptions are injected, not sampled from real mistakes, so the failure modes in here are the ones we chose to plant. This settles that a blocked graph does not run. It says nothing about how often an engineer writes one, and a paper quoting it as a base rate would be quoting it wrong.

Verify a design of your own

These rows describe neural network graphs that were checked by a deterministic verifier, and the same verifier is callable:

curl -X POST https://www.neurarch.com/api/v1/check \
  -H "Authorization: Bearer $NEURARCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d @graph.json

Cite

Neurarch. Verifier grounding study (264 graphs). https://neurarch.com/d/grounding-264.html

Downloads last month: 46