| { |
| "schema": 2, |
| "repo": "tphakala/Perch-v2", |
| "family": "perch", |
| "updated": "2026-07-24", |
| "family_meta": { |
| "display_name": "Google Perch v2", |
| "description": "Google Perch v2 multi-taxa classifier with approximately 14,795 species including birds, insects, amphibians, and mammals (scientific names only)", |
| "author": "Google Research", |
| "license": "Apache-2.0", |
| "commercial_use": true, |
| "upstream_url": "https://www.kaggle.com/models/google/bird-vocalization-classifier/tensorFlow2/perch_v2", |
| "species_count": 14795 |
| }, |
| "requirements_note": "requirements.min_ram_mb is a conservative, high-biased RAM floor: measured peak RSS + 15% where a benchmark exists, else a size-derived estimate biased high (a too-low floor risks OOM; a too-high floor only over-excludes on the smallest devices). Measured values will replace the estimates over time.", |
| "naming": "legacy names, frozen; new files follow <family>-<upstream>[-<region>]-<variant>-b<build>.<ext>", |
| "selection": { |
| "x86-64/onnxruntime": "perch-v2-fp32-no-dft.onnx", |
| "x86-64/openvino-cpu": "perch-v2-fp32-no-dft.onnx", |
| "aarch64-a76/openvino": "perch-v2-fp32-no-dft.onnx", |
| "aarch64/onnxruntime": "perch-v2-int8-arm.onnx", |
| "cuda": "perch-v2-fp32-no-dft.onnx" |
| }, |
| "notes": "perch-v2-fp32.onnx / perch_v2.onnx keep the in-graph DFT and therefore do NOT run on OpenVINO; use the no-dft build there. The int8-arm build was quantized from the with-DFT graph, so it is ONNX Runtime only (verified). The hyphenated and underscored names are duplicate content kept for backward compatibility.", |
| "models": [ |
| { |
| "path": "perch-v2-fp32.onnx", |
| "family": "perch", |
| "upstream_version": "2", |
| "upstream_source": "https://www.kaggle.com/models/google/bird-vocalization-classifier/", |
| "build": 1, |
| "legacy": true, |
| "variant": "fp32", |
| "precision": "fp32", |
| "region": null, |
| "classes": 14795, |
| "labels": "labels.txt", |
| "input": { |
| "sample_rate": 32000, |
| "window_seconds": 5, |
| "samples": 160000, |
| "shape": [ |
| "batch", |
| 160000 |
| ], |
| "dtype": "float32" |
| }, |
| "backends": { |
| "onnxruntime-cpu": { |
| "supported": true, |
| "recommended": true |
| }, |
| "openvino-cpu": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "openvino-gpu": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "cuda": { |
| "supported": true, |
| "recommended": true |
| }, |
| "tensorrt": { |
| "supported": true, |
| "recommended": false |
| } |
| }, |
| "backend_support_verified": false, |
| "sha256": "bf0c8467a924cb074663970ca4a0ab1e143602121930209657d0dff5d5cefa1f", |
| "size_bytes": 409148616, |
| "superseded_by": null, |
| "requirements": { |
| "min_ram_mb": 700 |
| } |
| }, |
| { |
| "path": "perch_v2.onnx", |
| "family": "perch", |
| "upstream_version": "2", |
| "upstream_source": "https://www.kaggle.com/models/google/bird-vocalization-classifier/", |
| "build": 1, |
| "legacy": true, |
| "variant": "fp32", |
| "precision": "fp32", |
| "region": null, |
| "classes": 14795, |
| "labels": "labels.txt", |
| "input": { |
| "sample_rate": 32000, |
| "window_seconds": 5, |
| "samples": 160000, |
| "shape": [ |
| "batch", |
| 160000 |
| ], |
| "dtype": "float32" |
| }, |
| "backends": { |
| "onnxruntime-cpu": { |
| "supported": true, |
| "recommended": true |
| }, |
| "openvino-cpu": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "openvino-gpu": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "cuda": { |
| "supported": true, |
| "recommended": true |
| }, |
| "tensorrt": { |
| "supported": true, |
| "recommended": false |
| } |
| }, |
| "backend_support_verified": false, |
| "sha256": "bf0c8467a924cb074663970ca4a0ab1e143602121930209657d0dff5d5cefa1f", |
| "size_bytes": 409148616, |
| "superseded_by": null, |
| "alias_of": "perch-v2-fp32.onnx", |
| "requirements": { |
| "min_ram_mb": 700 |
| } |
| }, |
| { |
| "path": "perch-v2-fp32-no-dft.onnx", |
| "family": "perch", |
| "upstream_version": "2", |
| "upstream_source": "https://www.kaggle.com/models/google/bird-vocalization-classifier/", |
| "build": 1, |
| "legacy": true, |
| "variant": "no-dft-fp32", |
| "precision": "fp32", |
| "region": null, |
| "classes": 14795, |
| "labels": "labels.txt", |
| "input": { |
| "sample_rate": 32000, |
| "window_seconds": 5, |
| "samples": 160000, |
| "shape": [ |
| "batch", |
| 160000 |
| ], |
| "dtype": "float32" |
| }, |
| "backends": { |
| "onnxruntime-cpu": { |
| "supported": true, |
| "recommended": false |
| }, |
| "openvino-cpu": { |
| "supported": true, |
| "recommended": true |
| }, |
| "openvino-gpu": { |
| "supported": true, |
| "recommended": true |
| }, |
| "cuda": { |
| "supported": true, |
| "recommended": true |
| }, |
| "tensorrt": { |
| "supported": true, |
| "recommended": false |
| } |
| }, |
| "backend_support_verified": true, |
| "sha256": "4dcf71c18a147198545944bb5149697e89e3ad2e16637fa8f0edf6d13035a017", |
| "size_bytes": 413350933, |
| "superseded_by": null, |
| "requirements": { |
| "min_ram_mb": 750 |
| } |
| }, |
| { |
| "path": "perch_v2_no_dft.onnx", |
| "family": "perch", |
| "upstream_version": "2", |
| "upstream_source": "https://www.kaggle.com/models/google/bird-vocalization-classifier/", |
| "build": 1, |
| "legacy": true, |
| "variant": "no-dft-fp32", |
| "precision": "fp32", |
| "region": null, |
| "classes": 14795, |
| "labels": "labels.txt", |
| "input": { |
| "sample_rate": 32000, |
| "window_seconds": 5, |
| "samples": 160000, |
| "shape": [ |
| "batch", |
| 160000 |
| ], |
| "dtype": "float32" |
| }, |
| "backends": { |
| "onnxruntime-cpu": { |
| "supported": true, |
| "recommended": false |
| }, |
| "openvino-cpu": { |
| "supported": true, |
| "recommended": true |
| }, |
| "openvino-gpu": { |
| "supported": true, |
| "recommended": true |
| }, |
| "cuda": { |
| "supported": true, |
| "recommended": true |
| }, |
| "tensorrt": { |
| "supported": true, |
| "recommended": false |
| } |
| }, |
| "backend_support_verified": true, |
| "sha256": "4dcf71c18a147198545944bb5149697e89e3ad2e16637fa8f0edf6d13035a017", |
| "size_bytes": 413350933, |
| "superseded_by": null, |
| "alias_of": "perch-v2-fp32-no-dft.onnx", |
| "requirements": { |
| "min_ram_mb": 750 |
| } |
| }, |
| { |
| "path": "perch-v2-int8-arm.onnx", |
| "family": "perch", |
| "upstream_version": "2", |
| "upstream_source": "https://www.kaggle.com/models/google/bird-vocalization-classifier/", |
| "build": 1, |
| "legacy": true, |
| "variant": "int8-arm", |
| "precision": "int8-arm", |
| "region": null, |
| "classes": 14795, |
| "labels": "labels.txt", |
| "input": { |
| "sample_rate": 32000, |
| "window_seconds": 5, |
| "samples": 160000, |
| "shape": [ |
| "batch", |
| 160000 |
| ], |
| "dtype": "float32" |
| }, |
| "backends": { |
| "onnxruntime-cpu": { |
| "supported": true, |
| "recommended": true |
| }, |
| "openvino-cpu": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "openvino-gpu": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "cuda": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "tensorrt": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| } |
| }, |
| "backend_support_verified": true, |
| "sha256": "ff32ca8c57954a86e6023a915d018dc7573cfc5567dd7314899d1c947cc6d5c5", |
| "size_bytes": 130856164, |
| "superseded_by": null, |
| "requirements": { |
| "min_ram_mb": 350 |
| } |
| }, |
| { |
| "path": "perch_v2_int8_arm.onnx", |
| "family": "perch", |
| "upstream_version": "2", |
| "upstream_source": "https://www.kaggle.com/models/google/bird-vocalization-classifier/", |
| "build": 1, |
| "legacy": true, |
| "variant": "int8-arm", |
| "precision": "int8-arm", |
| "region": null, |
| "classes": 14795, |
| "labels": "labels.txt", |
| "input": { |
| "sample_rate": 32000, |
| "window_seconds": 5, |
| "samples": 160000, |
| "shape": [ |
| "batch", |
| 160000 |
| ], |
| "dtype": "float32" |
| }, |
| "backends": { |
| "onnxruntime-cpu": { |
| "supported": true, |
| "recommended": true |
| }, |
| "openvino-cpu": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "openvino-gpu": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "cuda": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| }, |
| "tensorrt": { |
| "supported": false, |
| "recommended": false, |
| "note": "contains an in-graph DFT op that OpenVINO cannot translate; verified to fail compilation on OpenVINO CPU" |
| } |
| }, |
| "backend_support_verified": true, |
| "sha256": "ff32ca8c57954a86e6023a915d018dc7573cfc5567dd7314899d1c947cc6d5c5", |
| "size_bytes": 130856164, |
| "superseded_by": null, |
| "alias_of": "perch-v2-int8-arm.onnx", |
| "requirements": { |
| "min_ram_mb": 350 |
| } |
| } |
| ] |
| } |
|
|
