Modelpublic

skillsafe-ai/slimsam-77-uniform-int8

sourceHugging Faceapache-2.0updated 4d agoView on Hugging Face

0likes31downloads

Model Card

SlimSAM-77 uniform segment anything (int8)

Browser-ready import artifacts for image-segmentation, produced by SkillSafe's reproducible converter (`models/` in skillsafe.ai) from a pinned upstream source. Every byte here is derivable from that source plus the recipe below; nothing was edited by hand.

Provenance


Upstream	https://huggingface.co/Xenova/slimsam-77-uniform/tree/5850ab45f587c112167512ffef949107115e26a0
Upstream SHA-256 / commit	`5850ab45f587c112167512ffef949107115e26a0`
Recipe	`recipes/slimsam-77-uniform-int8.yaml` — sha256 `33c47dabbe648c14c0c353240554658afdfceed2e07c50aa21bb3fa7832845da`
Toolchain	Python 3.12.13, torch 2.10.0, onnx 1.23.0, onnxruntime 1.30.0 on Darwin 25.6.0 arm64
Converted	2026-09-22T21:39:52+00:00

Files

file	class	size	SHA-256
`config.json`	bundle	0.00 MB	`6339884f168658d3ca6473b486973913fb33e84e625e06ae2dd7b4a808187419`
`onnx/prompt_encoder_mask_decoder_quantized.onnx`	registry (int8)	4.68 MB	`cb90b279f549d2cab7fd6e20c38522438c65d84bdcca3d2a764cff7d857fdce2`
`onnx/vision_encoder_quantized.onnx`	registry (int8)	8.47 MB	`cce23c7b2e5d4f330932738fb67ba518e04b0d99ccdd1cccd22a7da4e01f2971`
`preprocessor_config.json`	bundle	0.00 MB	`225545a743c654e3c495ec6f545a0eaba57c8ba3fbbd8483b3cb1c0fc58db517`

registry files are parameter files served from models.skillsafe.ai once vetted; bundle files ship inside an app; registry-shared is a runtime library reused by every model of the same architecture.

Verification

Imported as published upstream (no conversion). Each file is pinned by SHA-256 to its source; every ONNX file passed onnx.checker and a CPU smoke run under onnxruntime with zero-filled inputs at the declared shapes:

file	inputs	outputs	ms
`onnx/prompt_encoder_mask_decoder_quantized.onnx`	inputpoints[1, 1, 1, 2], inputlabels[1, 1, 1], imageembeddings[1, 256, 64, 64], imagepositional_embeddings[1, 256, 64, 64]	iouscores[1, 1, 3], predmasks[1, 1, 3, 256, 256]	11.9
`onnx/vision_encoder_quantized.onnx`	pixel_values[1, 3, 1024, 1024]	imageembeddings[1, 256, 64, 64], imagepositional_embeddings[1, 256, 64, 64]	388.4

Use in the browser

import * as ort from "onnxruntime-web";
const session = await ort.InferenceSession.create("https://huggingface.co/skillsafe-ai/slimsam-77-uniform-int8/resolve/main/onnx/prompt_encoder_mask_decoder_quantized.onnx", { executionProviders: ["webgpu", "wasm"] });

Contract (onnx/prompt_encoder_mask_decoder_quantized.onnx): input input_points float32 ['batch_size', 'point_batch_size', 'nb_points_per_image', 2], input_labels int64 ['batch_size', 'point_batch_size', 'nb_points_per_image'], image_embeddings float32 ['batch_size', 256, 64, 64], image_positional_embeddings float32 ['batch_size', 256, 64, 64] → output iou_scores float32 ['batch_size', 'point_batch_size', 3], pred_masks float32 ['batch_size', 'point_batch_size', 3, 256, 256]. Opset 13.

Licence and attribution

SlimSAM: 0.1% Data Makes Segment Anything Slim (Chen et al., NeurIPS 2024), a structural pruning of Meta's Segment Anything ViT-B, Apache License 2.0. ONNX export published by Xenova (https://huggingface.co/Xenova/slimsam-77-uniform).

Licence: Apache-2.0 — notice: https://github.com/czg1225/SlimSAM/blob/main/LICENSE. The conversion recipe and this model card are part of the SkillSafe repository and carry its licence; the weights remain under the upstream licence above.

The full manifest.json in this repo records the recipe, sources, toolchain (including the uv.lock hash) and per-file verification numbers.