Datasets:

Shanmuk4622
/

msc-cifar100

sample_idx int64 0 50k	forget_events int32 0 86	ever_correct bool 2 classes	el2n float32 0 1.41 ⌀	unforgettable bool 2 classes
0	28	true	0.452721	false
1	0	true	0.002747	true
2	32	true	0.13591	false
3	26	true	0.879346	false
4	5	true	0.022034	false
5	36	true	1.360101	false
6	24	true	0.260248	false
7	9	true	0.432419	false
8	21	true	0.40303	false
9	19	true	0.232125	false
10	4	true	0.189115	false
11	31	true	0.816132	false
12	2	true	0.18261	false
13	40	true	0.147988	false
14	2	true	0.322631	false
15	37	true	1.0696	false
16	5	true	0.115399	false
17	40	true	0.675364	false
18	7	true	0.219234	false
19	0	true	0.126509	true
20	5	true	1.063041	false
21	35	true	1.018327	false
22	37	true	0.934044	false
23	24	true	0.62749	false
24	32	true	0.055535	false
25	12	true	0.591011	false
26	42	true	1.127939	false
27	21	true	0.415851	false
28	41	true	0.630348	false
29	7	true	0.984912	false
30	2	true	0.184602	false
31	16	true	0.021176	false
32	26	true	0.352804	false
33	29	true	1.03717	false
34	12	true	0.08832	false
35	29	true	0.679644	false
36	33	true	0.540707	false
37	19	true	1.0418	false
38	13	true	0.419033	false
39	5	true	0.129835	false
40	35	true	1.063214	false
41	11	true	0.068141	false
42	5	true	0.382026	false
43	11	true	0.002759	false
44	10	true	0.708646	false
45	12	true	0.114066	false
46	30	true	0.784891	false
47	23	true	0.21369	false
48	9	true	1.379132	false
49	29	true	0.968989	false
50	25	true	0.547224	false
51	35	true	1.049061	false
52	8	true	0.235764	false
53	20	true	0.519568	false
54	3	true	0.087212	false
55	37	true	0.867328	false
56	31	true	0.268573	false
57	0	true	0.001068	true
58	26	true	0.832861	false
59	42	true	0.563131	false
60	29	true	0.139431	false
61	29	true	0.94365	false
62	18	true	0.519998	false
63	25	true	1.157752	false
64	15	true	1.198787	false
65	34	true	0.677259	false
66	37	true	0.879242	false
67	11	true	0.355402	false
68	36	true	0.43475	false
69	34	true	0.163445	false
70	19	true	0.61804	false
71	40	true	0.815185	false
72	6	true	0.14877	false
73	18	true	0.768833	false
74	8	true	0.004377	false
75	21	true	0.671154	false
76	12	true	0.453153	false
77	21	true	0.609947	false
78	2	true	0.008255	false
79	7	true	0.177203	false
80	14	true	0.064684	false
81	24	true	1.016114	false
82	4	true	0.25589	false
83	12	true	1.058184	false
84	3	true	0.202077	false
85	38	true	0.889215	false
86	9	true	1.142873	false
87	10	true	0.384633	false
88	3	true	0.067684	false
89	17	true	0.902272	false
90	34	true	0.868444	false
91	4	true	1.033213	false
92	14	true	0.674693	false
93	7	true	0.385542	false
94	33	true	0.739241	false
95	24	true	1.308898	false
96	7	true	0.046554	false
97	4	true	0.55596	false
98	22	true	0.820617	false
99	19	true	0.924641	false

End of preview. Expand in Data Studio

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

MSC — Minimum Sufficient Compute

Artifacts for Is Compute Difficulty Architecture-Agnostic? Measuring and Distilling Per-Sample Minimum Sufficient Computation.

Generated 2026-08-06T03:56:37Z by msc_lib v1.0.0.

Repositories

Shanmuk4622/msc-cifar100 — everything, one folder per run

What MSC is

The smallest cost-normalised configuration at which a network's decision has stably settled to its full-compute decision, defined uniformly over depth, resolution and precision reduction. Stability means the decision agrees at that budget and every larger one — predictions under compute reduction are not monotone, so a naive minimum records an accident rather than a property.

Compute grid

depth: exits at [0.2, 0.4, 0.6, 0.8, 1.0] of network depth
resolution: [16, 20, 24, 28, 32] px, measured natively AND via a downsample-upsample proxy (the proxy cost model is labelled idealised)
precision: ['int4', 'int6', 'int8', 'fp16', 'fp32'], simulated by fake quantisation; cost priced analytically as bits/32, never as measured latency
confidence thresholds: tau in [0.0, 0.1, 0.2, 0.3, 0.5] — all results are tau-curves

Per-image table schema

sample_idx, label,
pred_d1..d5    top1p_d1..d5    top2p_d1..d5     depth
pred_rn1..rn5  top1p_rn1..rn5  top2p_rn1..rn5   resolution (native)
pred_rp1..rp5  top1p_rp1..rp5  top2p_rp1..rp5   resolution (proxy)
pred_q1..q5    top1p_q1..q5    top2p_q1..q5     precision
msp, margin, entropy, ce_loss, el2n, forget_events, pred_depth
sample_order_hash, run_id, split

Every table carries sample_order_hash. Tables whose hashes differ are not row-aligned and must not be correlated.

Telemetry recorded per epoch

Losses and accuracies; learning rate per group; gradient norm mean/max/p95; gradient-clip hit rate; weight norm; update-to-weight ratio; AMP scale; NaN/Inf batch count; epoch/train/eval time; dataload vs compute split; step-time p50/p90/p99; throughput; VRAM allocated/reserved/peak; GPU utilisation and temperature; CPU and RAM; free disk; energy in J/kWh and CO2 per epoch and cumulative. Plus raw power samples at 10 Hz, system samples at 1 Hz, and a downsampled per-step trace.

Reproducibility

config.yaml frozen at run start, sha256-hashed, asserted on resume
checkpoints carry optimizer, scheduler, AMP scaler and all four RNG streams
3 seeds per headline number, mean +/- std
every artifact mapped to a run_id in paper/provenance.csv
work split across accounts by a deterministic cost-balanced scheduler; each run records which worker produced it

Atlas results

arch	seed	acc_pct	reference_accuracy	params_M	GFLOPs
convnext_femto	1	62.67	nan	4.87	0.126
convnext_femto	2	62.6	nan	4.87	0.126
convnext_femto	3	61.84	nan	4.87	0.126
mixer_nano	1	60.23	nan	2.5	0.343
mixer_nano	2	60.32	nan	2.5	0.343
mixer_nano	3	60.72	nan	2.5	0.343
mobilenetv2	1	70.1	64.6	2.35	0.183
mobilenetv2	2	69.89	64.6	2.35	0.183
mobilenetv2	3	70.31	64.6	2.35	0.183
shufflenetv2	1	71.93	70.5	1.36	0.092
shufflenetv2	2	71.54	70.5	1.36	0.092
shufflenetv2	3	71.81	70.5	1.36	0.092
resnet110	1	74.31	74.31	1.74	0.511
resnet110	2	74.57	74.31	1.74	0.511
resnet110	3	74.26	74.31	1.74	0.511
resnet20	1	70.25	69.06	0.28	0.082
resnet20	2	70.36	69.06	0.28	0.082
resnet20	3	69.78	69.06	0.28	0.082
resnet32x4	1	79.59	79.42	7.42	2.149
resnet32x4	1	79.54	79.42	7.42	2.149
resnet32x4	2	79.63	79.42	7.42	2.149
resnet32x4	2	80.03	79.42	7.42	2.149
resnet32x4	3	79.46	79.42	7.42	2.149
resnet56	1	73.88	72.34	0.86	0.254
resnet56	2	73.35	72.34	0.86	0.254
resnet56	3	73.85	72.34	0.86	0.254
resnet8x4	1	73.35	72.5	1.22	0.334
resnet8x4	2	73.39	72.5	1.22	0.334
resnet8x4	3	73.04	72.5	1.22	0.334
vgg13	1	75.7	74.64	9.46	0.458
vgg13	2	75.65	74.64	9.46	0.458
vgg13	3	75.75	74.64	9.46	0.458
vgg8	1	71.56	70.36	3.96	0.136
vgg8	2	71.73	70.36	3.96	0.136
vgg8	3	71.61	70.36	3.96	0.136
vit_tiny	1	59.33	nan	5.38	0.694
vit_tiny	2	59.13	nan	5.38	0.694
vit_tiny	3	60.56	nan	5.38	0.694
wrn_16_2	1	73.64	73.26	0.7	0.203
wrn_16_2	2	73.79	73.26	0.7	0.203
wrn_16_2	3	74.27	73.26	0.7	0.203
wrn_40_1	1	72.41	71.98	0.57	0.168
wrn_40_1	2	72.54	71.98	0.57	0.168
wrn_40_1	3	72.28	71.98	0.57	0.168
wrn_40_2	1	76.89	75.61	2.26	0.658
wrn_40_2	1	76.06	75.61	2.26	0.658
wrn_40_2	2	76.72	75.61	2.26	0.658
wrn_40_2	2	76.62	75.61	2.26	0.658
wrn_40_2	3	76.16	75.61	2.26	0.658

Limitations

Per-image routing gives no wall-clock speedup under batched inference unless the batch is split by route. The deployment claim is scoped to batch-1 / edge / streaming.
INT4 and INT6 are simulated; no T4 kernel exists to time them.
The resolution proxy runs at 32 px; its cost is an idealised model.
Risk control is calibrated at epsilon=0.03 on a 5,000-image holdout, because epsilon=0.01 would need ~14,979 calibration images and the CIFAR-100 test set has 10,000.
Energy is measurement methodology, not a contribution.
T4-only hardware; CIFAR-100 scale.

Downloads last month: 1,010