--- library_name: mobilint license: apache-2.0 base_model: - timm/caformer_m36.sail_in1k base_model_relation: quantized pipeline_tag: image-classification tags: - mobilint --- # CAFormer_M36 `aries-rb/` holds the MXQ reproduced from the published arxiv folder (https://github.com/mobilint/mblt-model-arxiv/tree/main/models/image_classification/CAFormer_M36). It was compiled by qbcompiler 1.4.0 in the image `mobilint/qbcompiler:1.4-cuda12.8.1-ubuntu22.04` (sha256:7910f1661605b82eb9616f6ac99fd6461dbb865dcadaa810cce22c638ceffd44) with `--inference-scheme all`, so it supports the core mode(s) `single`, `multi`, `global4`, `global8` only. It was evaluated with qbruntime 1.5.0 in `async` inference mode on imagenet (50000 samples): top1_accuracy 0.83948, top5_accuracy 0.96768. Only the evaluated core mode was measured. SHA-256: `14455c59f6fd466c3dec6c9adab476799f42663bbaa878dea2eab4b6ce422fe0` (96012342 bytes). Verify a download with `sha256sum .mxq` and compare the digest with this value and with `aries-rb/metadata.json`. aries-rb/best_result.json contains legacy build values and is retained for mblt-vision-python compile defaults. The legacy aries/ directory was removed; mblt-model-zoo <= 2.12 must pin an older revision. ```json { "schema_version": 2, "model": "CAFormer_M36", "variant": "c0923-asym", "qbcompiler": "1.4.0", "qbruntime": "1.5.0", "compile": { "image_tag": "1.4-cuda12.8.1-ubuntu22.04", "image_digest": "sha256:7910f1661605b82eb9616f6ac99fd6461dbb865dcadaa810cce22c638ceffd44", "inference_scheme": "all", "device": "gpu" }, "evaluation": { "dataset": "imagenet", "sample_count": 50000, "primary": { "top1_accuracy": 0.83948 }, "secondary": { "top5_accuracy": 0.96768 }, "protocol": null, "eval_data_identity": { "annotation_sha256": "f00550c10ff1d4110abc683786b6cc09d5082dadd3cee1568f5cfa1dad962960", "sample_list_sha256": "54446819dced83110352e5a0361ff05b19c331c5a9802dcb0222e6d1e44a0547" }, "identity_audit": "verified", "inference_mode": "async", "image_tag": "1.4-cpu-ubuntu22.04", "image_digest": "sha256:62b5bc714571dffd94fb561275767ca80d15337a433630a3ae7936305d531a7b" }, "retention_percent": { "top1_accuracy": 98.50508084766845 }, "content_sha256": "6eb720bea1b6e097d0a42aa6a9c13029da4a7c9545dc109527f49fe201b2adbb", "attempt_id": "CAFormer_M36-jinman-coreall-65f5484d28b8", "superseded_attempt_ids": [], "sha256": "14455c59f6fd466c3dec6c9adab476799f42663bbaa878dea2eab4b6ce422fe0", "format": "MXQv7", "supported_modes": [ "single", "multi", "global4", "global8" ], "size": 96012342, "ops_commits": { "selection": "417dd6508f5a1f5fce40ca68dbdf1cd0e9e2ccf5" }, "arxiv_folder": "https://github.com/mobilint/mblt-model-arxiv/tree/main/models/image_classification/CAFormer_M36", "publication": { "status": "public", "reason": "approved" } } ```