Modelpublic

FasterAI-Labs/pilot-resnet18-imagenette

sourceHugging Facebsd-3-clauseupdated 13d agoView on Hugging Face

0likes28downloads

Model Card

resnet18 — Imagenette pilot

Imagenette validation, 3925 images, one run. Every published file is scored after reload; the reference is the fine-tuned source.

Criteria

Top-1 on the evaluation set named above, size on disk, peak live activations and multiply-accumulates — the last two for one image of (3, 160, 160) at batch 1 — each against the fine-tuned resnet18.

pruned FP32 — load with fastermodels.load(repo) (`model.safetensors`)

criterion	reference	this artifact	gap
top-1	93.3 %	91.4 %	-1.9 pt [-2.7, -1.1]
size	44.8 MB, 11.2 M params	21.1 MB, 5.3 M params	-52.8 %
memory	3.3 MB	2.0 MB	-37.5 %
MACs	925.3 M	425.8 M	-54.0 %

INT8 TorchScript — load with fastermodels.load(repo) or torch.jit.load (`model.torchscript.pt`)

criterion	reference	this artifact	gap
top-1	93.3 %	91.3 %	-2.0 pt [-2.8, -1.2]
size	44.8 MB, 11.2 M params	5.4 MB, n/a params	-88.0 %
memory	3.3 MB	0.4 MB	-88.3 %
MACs	925.3 M	425.8 M	-54.0 %

INT8 ONNX — load with any ONNX runtime, scored here through onnxruntime (`model.onnx`)

criterion	reference	this artifact	gap
top-1	93.3 %	91.2 %	-2.1 pt [-2.9, -1.3]
size	44.8 MB, 11.2 M params	5.4 MB, n/a params	-88.0 %
memory	3.3 MB	0.4 MB	-88.3 %
MACs	925.3 M	425.8 M	-54.0 %

Gaps are measured on the same images as the reference; brackets give the 95 % interval.

Latency

non mesurée

Provenance

—Source model: torchvision.models.resnet18 (IMAGENET1K_V1)
—fasterai: 0.4.0
—fastermodels: 0.1.0
—torch: 2.9.1+cu128
—measured_on: 2026-09-11
—sourcestatehash: b81b9caef3c9

Publication checks: 10/10 passed.