Sync model repo (text/metadata)

README.md

@@ -0,0 +1,160 @@
+---
+license: bsd-3-clause
+pipeline_tag: image-classification
+library_name: executorch
+base_model: timm/efficientnet_b4.ra2_in1k
+base_model_relation: quantized
+tags:
+- arm
+- edge-linux
+- executorch
+- xnnpack
+- int8
+- imagenet-1k
+---
+
+# efficientnet_b4.ra2_in1k optimized for Arm-based Edge Linux
+
+efficientnet_b4.ra2_in1k image classification model, optimized with W8A8 dynamic post-training quantization and exported as an ExecuTorch program for the XNNPACK CPU backend on Arm-based platforms.
+
+This model was optimized using [Arm AI Model Optimizer](https://developer.arm.com/dev-docs/ai-model-optimizer/).
+
+## Summary
+
+This repository contains an optimized version of timm/efficientnet_b4.ra2_in1k for image classification. The model is provided as an ExecuTorch `.pte` file using the XNNPACK with KleidiAI delegate, targeting Edge Linux systems. Precision: INT8 symmetric per-channel weights, INT8 dynamic activations.
+
+**Key results**
+
+| Area | Result |
+|---|---|
+| Model format | ExecuTorch (.pte) |
+| Target device class | Edge Linux |
+| Reference device | Raspberry Pi 5 (16GB) (Cortex-A76, 4 cores) |
+| Primary performance result | 1.09x lower p50 latency (467.94 ms to 431.22 ms) |
+| Accuracy result | Top-1 accuracy 82.74% (-0.32 pp vs. baseline) |
+| Size / memory result | 3.66x smaller model (79.61 MB to 21.77 MB), 132.91 MB peak memory |
+
+## Original model
+
+| Field | Value |
+|---|---|
+| Original model | timm/efficientnet_b4.ra2_in1k |
+| Original source | Hugging Face |
+| Original developer | timm |
+| Original model card | [timm/efficientnet_b4.ra2_in1k](https://huggingface.co/timm/efficientnet_b4.ra2_in1k) |
+| Original license | [BSD-3-Clause](https://opensource.org/license/bsd-3-clause) |
+
+## Model files
+
+| File | Description |
+|---|---|
+| `model.pte` | Optimized INT8 ExecuTorch program for deployment |
+| `metadata.yaml` | Optimized model metadata |
+| `benchmarks/` | Baseline and optimized benchmark records |
+
+## Performance
+
+Performance was measured on the reference configuration below. Results do not guarantee identical performance on every Arm-based system.
+
+**Reference configuration**
+
+| Field | Value |
+|---|---|
+| Device / platform | Raspberry Pi 5 (16GB) |
+| CPU / accelerator | Cortex-A76, 4 cores, 2.4 GHz |
+| OS | Raspberry Pi OS Bookworm |
+| Runtime | ExecuTorch 1.1.0 |
+| Backend / delegate | XNNPACK with KleidiAI, 4 threads |
+| Batch size | 1 |
+| Input resolution | 384x384 |
+| Precision | INT8 symmetric per-channel weights, INT8 dynamic activations |
+| Runs | 10 measured |
+
+**Performance results**
+
+| Metric | Original / baseline | Optimized | Improvement |
+|---|---:|---:|---:|
+| p50 end-to-end latency | 467.94 ms | 431.22 ms | 1.09x |
+| Throughput | - | 2.32 frames/sec | - |
+| Peak memory | 188.92 MB | 132.91 MB | 1.42x less |
+| Model size | 79.61 MB | 21.77 MB | 3.66x smaller |
+
+## Accuracy
+
+**Evaluation setup**
+
+| Field | Value |
+|---|---|
+| Dataset | ImageNet-1k |
+| Split | val |
+| Number of samples | 5,000 |
+| Evaluation runtime | ExecuTorch |
+
+**Accuracy results**
+
+| Metric | Original / baseline | Optimized | Change |
+|---|---:|---:|---:|
+| Top-1 accuracy | 83.06% | 82.74% | -0.32 pp |
+| Top-5 accuracy | - | 96.72% | - |
+
+Users should re-evaluate the model on their own data before production use.
+
+## Optimization approach
+
+| Optimization area | Applied? | Notes |
+|---|---|---|
+| Model conversion | Yes | PyTorch model exported to ExecuTorch (.pte) |
+| Quantization | Yes | INT8 symmetric per-channel weights, INT8 dynamic activations. Calibrated on 1,000 random samples (imagenet1k_calib_seed42_n1000) |
+| Runtime/backend selection | Yes | XNNPACK with KleidiAI, 4 threads |
+| Accuracy validation | Yes | ImageNet-1k, 5000 samples |
+| Performance validation | Yes | Measured on Raspberry Pi 5 (16GB) |
+
+## Intended use
+
+This model is intended for developers evaluating image classification workloads on Arm-based Edge Linux platforms. It is suitable as a reference implementation for benchmarking, prototyping, and integration exploration.
+
+## Limitations
+
+- Performance depends on the target device, runtime version, backend support, memory configuration, and system load.
+- Accuracy was measured on a 5,000-image subset of the ImageNet-1k validation set, not the full 50,000 images.
+- This repository is not a replacement for the original model documentation.
+
+## About this version
+
+Original Model:
+timm/efficientnet_b4.ra2_in1k by timm - [Repository](https://huggingface.co/timm/efficientnet_b4.ra2_in1k)
+
+Optimization/conversion:
+W8A8 dynamic post-training quantization and ExecuTorch `.pte` export with XNNPACK + KleidiAI backend support for execution on Arm-based platforms.
+
+Converted/optimized by:
+Nota AI, using Arm AI Model Optimizer
+
+License:
+The Original Model and the Optimized Model are subject to [BSD-3-Clause](https://opensource.org/license/bsd-3-clause).
+
+This repository contains a converted or optimized version of the Original Model (the "Optimized Model"). The Original Model has been converted or optimized as described above for execution on Arm-based platforms.
+
+No retraining or fine-tuning of the Original Model was performed as part of the conversion or optimization. The conversion or optimization was not intended to change the Original Model's behavior or intended use.
+
+### Original Model and Documentation
+
+For information about the Original Model, including its development, training data, intended uses, limitations and other relevant information, please refer to the [Original Model repository](https://huggingface.co/timm/efficientnet_b4.ra2_in1k). Information in that repository was provided by the original developer or other third parties and, unless expressly stated otherwise, has not been independently verified by Arm.
+
+### Licenses and Third-Party Terms
+
+Use of the Original Model and the Optimized Model is subject to the applicable licenses, usage restrictions and other terms identified above and in the relevant repositories. Publication of the Optimized Model does not grant any rights beyond those provided under the applicable license terms.
+
+You are responsible for reviewing those terms and ensuring that your use of the Original Model and the Optimized Model is permitted.
+
+### Purpose of this Release
+
+The Optimized Model is provided as a reference implementation to demonstrate and evaluate execution and performance on Arm-based systems. It is not a production-ready or supported solution.
+
+Arm's publication of the Optimized Model does not constitute an endorsement or certification of the Original Model or a representation that the Optimized Model is suitable for production use or any particular purpose.
+
+To the fullest extent permitted by applicable law (i) the Optimized Model is provided "as is." Arm makes no representations or warranties that the Original Model, the Optimized Model or their outputs are accurate, safe, secure, non-infringing, legally compliant, suitable for production use or fit for any particular purpose; and (ii) Arm will not be liable for any loss or damage arising from or in connection with the Optimized Model, its use or its outputs.
+
+You are responsible for independently evaluating the Optimized Model, its outputs and its suitability for your intended use, including compliance with applicable legal, regulatory, safety and security requirements.
+
+Arm does not commit to provide ongoing support, maintenance or updates for the Optimized Model. Any use of or reliance on the Optimized Model or its outputs is at your own risk.

benchmarks/efficientnet-b4-raspberrypi5-fp32.yaml

@@ -0,0 +1,30 @@
+version: 1.0.0
+report_type: vision-classification
+profile: Baseline
+created_at: '2026-06-21T16:02:16.993793Z'
+context:
+  target_ref: raspberry-pi-5
+  runtime:
+    name: executorch
+    execution_backend: cpu
+    version: 1.1.0
+    config:
+      n_threads: 4
+      optimisations:
+      - XNNPACK
+      - KleidiAI
+  dataset:
+    name: ImageNet-1k
+    sample_count: 5000
+    slice: val
+    reference_url: https://www.image-net.org/
+  benchmark:
+    batch_size: 1
+    num_runs: 10
+    input_resolution: 384x384
+performance:
+  end_to_end_latency_ms:
+    p50: 467.94
+  peak_memory_mb: 188.92
+accuracy:
+  top_1_accuracy_pct: 83.06

benchmarks/efficientnet-b4-raspberrypi5-int8.yaml

@@ -1,27 +1,8 @@
 version: 1.0.0
 report_type: vision-classification
-created_at: '2026-06-21T16:02:16.993793Z'
 profile: Arm-Optimized
+created_at: '2026-06-21T16:02:16.993793Z'
 context:
-  model:
-    id: arm/efficientnet-b4-arm-optimized
-    base_model_id: TBD/efficientnet-b4
-    filename: model.pte
-    profile: Arm-Optimized
-    weight_dtype: int8
-    model_size_mb: 24.29
-    format: pte
-    quantization:
-      method: PTQ-dynamic
-      weight_bits: 8
-      activation_bits: 8
-      symmetric: true
-      mode: dynamic
-      weight_granularity: per-channel
-    calibration:
-      dataset_name: imagenet1k_calib_seed42_n1000
-      sample_count: 1000
-      selection: random
   target_ref: raspberry-pi-5
   runtime:
     name: executorch

metadata.yaml

@@ -4,10 +4,10 @@ id: Arm/efficientnet-b4-int8-xnnpack-executorch-raspberrypi5
 task_type: image-classification
 title: EfficientNet-B4 INT8
 filename: model.pte
-base_model_id: TBD/efficientnet-b4
+base_model_id: timm/efficientnet_b4.ra2_in1k
 weight_dtype: int8
 profile: Arm-Optimized
-model_size_mb: 24.29
+model_size_mb: 21.77
 format: pte
 quantization:
   method: PTQ-dynamic

np_recipe/calibration.json

@@ -1,11 +0,0 @@
-{
-  "datasets": [
-    {
-      "ref_id": "595edc1f-9919-4027-b23d-1436459990fa",
-      "dataset_name": "imagenet1k_calib_seed42_n1000",
-      "split": "CALIBRATION",
-      "num_samples": 1000,
-      "original_dataset_alias": "imagenet1k"
-    }
-  ]
-}

np_recipe/evaluation.json

@@ -1,11 +0,0 @@
-{
-  "datasets": [
-    {
-      "ref_id": "384e7ec8-2e03-4f88-9c7f-f7a248ae8403",
-      "dataset_name": "imagenet1k_eval_seed42_n5000",
-      "split": "EVALUATION",
-      "num_samples": 5000,
-      "original_dataset_alias": "imagenet1k"
-    }
-  ]
-}

np_recipe/project.json

@@ -1,24 +0,0 @@
-{
-  "project_id": "8c53bd22-3f4b-4b83-b9c3-2063e78c2c29",
-  "name": "np_hpo_97c660c913",
-  "target": {
-    "device": "raspberry_pi_5_16gb",
-    "runtime": "executorch",
-    "backend": "xnnpack"
-  },
-  "gpu_device": "cuda:1",
-  "base_model": {
-    "model_id": "f1f45b7b-c443-4b94-86c2-f3e5f0bbd7aa",
-    "alias": "efficientnet_b4_pt2_pytorch_static_fp32",
-    "modality": "CV",
-    "task": "CLASSIFICATION"
-  },
-  "calibration": {
-    "dataset_id": "595edc1f-9919-4027-b23d-1436459990fa",
-    "dataset_name": "imagenet1k_calib_seed42_n1000"
-  },
-  "evaluation": {
-    "dataset_id": "384e7ec8-2e03-4f88-9c7f-f7a248ae8403",
-    "dataset_name": "imagenet1k_eval_seed42_n5000"
-  }
-}

np_recipe/run.detail.yaml

@@ -1,95 +0,0 @@
-detail:
-  batch_size: 1
-  gpu_device: cuda:1
-  graph_optimize[0]:
-    pattern_handlers:
-    - PatternHandlerAdjustAdd
-    - PatternHandlerAdjustNonContiguousConstantMathInput
-    - PatternHandlerCanonicalizeConvolutionDefault
-    - PatternHandlerCombineAddRelu
-    - PatternHandlerCombineConvBN
-    - PatternHandlerCombineConvBNRelu
-    - PatternHandlerCombineConvRelu
-    - PatternHandlerCombineConvTransposeBN
-    - PatternHandlerCombineConvTransposeBNRelu
-    - PatternHandlerCombineConvTransposeRelu
-    - PatternHandlerCombineLinearRelu
-    - PatternHandlerCombineMulRelu
-    - PatternHandlerCommuteBNBeforeJoinOp
-    - PatternHandlerCommuteBNBeforeLinearOp
-    - PatternHandlerComposeErfFormulaIntoGelu
-    - PatternHandlerComposeTanhFormulaIntoGelu
-    - PatternHandlerDecomposeMaxDimKeepdimFalse
-    - PatternHandlerDecomposePReluToReluMath
-    - PatternHandlerFuseAffineInputToLinear
-    - PatternHandlerFuseBNToConv
-    - PatternHandlerFuseConsecutiveBatchNorm
-    - PatternHandlerFuseConsecutiveCast
-    - PatternHandlerFuseConsecutiveConcat
-    - PatternHandlerFuseConsecutiveIdempotent
-    - PatternHandlerFuseConsecutiveLogSoftmax
-    - PatternHandlerFuseConsecutiveReshapeTranspose
-    - PatternHandlerFuseConsecutiveSlice
-    - PatternHandlerFuseConsecutiveUnsqueeze
-    - PatternHandlerFuseConstantPath
-    - PatternHandlerFuseInputScaleToConv
-    - PatternHandlerFuseMathToConv
-    - PatternHandlerFuseMathToLinear
-    - PatternHandlerFusePadToConv
-    - PatternHandlerFusePadToPool
-    - PatternHandlerFuseReluPostfixToClamp
-    - PatternHandlerFuseReluPrefixToClamp
-    - PatternHandlerFuseMultiBranchReshapeTranspose
-    - PatternHandlerFuseQKV
-    - PatternHandlerFuseSplitConvConcat
-    - PatternHandlerFuseUnsqueezeToReduce
-    - PatternHandlerNormalizeSpatialMeanKeepdimFalse
-    - PatternHandlerRemoveGraphAssertsPass
-    - PatternHandlerRemoveBatchNormTracking
-    - PatternHandlerRemoveCommonSubexpression
-    - PatternHandlerRemoveDeadend
-    - PatternHandlerRemoveDisabledAutocastWrapper
-    - PatternHandlerRemoveDropout
-    - PatternHandlerRemoveMonotoneBeforeRanking
-    - PatternHandlerRemoveNoOpCast
-    - PatternHandlerRemoveNoOpConcat
-    - PatternHandlerRemoveNoOpExpand
-    - PatternHandlerRemoveNoOpFlatten
-    - PatternHandlerRemoveNoOpMath
-    - PatternHandlerRemoveNoOpPad
-    - PatternHandlerRemoveNoOpReshape
-    - PatternHandlerRemoveNoOpSlice
-    - PatternHandlerRemoveNoOpSplit
-    - PatternHandlerRemoveNoOpSqueeze
-    - PatternHandlerRemoveNoOpTranspose
-    - PatternHandlerRemovePositive
-    - PatternHandlerReplaceBatchNormWithConv
-    - PatternHandlerReplaceBitwiseRightShiftWithDiv
-    - PatternHandlerReplaceEinsumWithMatmul
-    - PatternHandlerReplaceGatherWithConv
-    - PatternHandlerReplaceGatherWithSliceReshape
-    - PatternHandlerReplaceInplaceAdd
-    - PatternHandlerReplaceInplaceHardswish
-    - PatternHandlerReplaceInplaceIndexPut
-    - PatternHandlerReplaceInplaceLeakyRelu
-    - PatternHandlerReplaceInplaceMul
-    - PatternHandlerReplaceInplaceRelu
-    - PatternHandlerReplaceInplaceSilu
-    - PatternHandlerReplaceMatmul2dWithLinear
-    - PatternHandlerReplaceMatmul3dWithLinear
-    - PatternHandlerReplaceMatmulWithLinear
-  graph_quantize[0]:
-    weight:
-      granularity: CHANNEL
-      calibration_method: MINMAX
-    activation:
-      granularity: TENSOR
-      calibration_method: MINMAX
-    calibration:
-      num_samples: 128
-  profile[0]:
-    num_executions: 10
-    timeout: 1800
-  evaluate[0]:
-    metrics:
-      ACCURACY: {}

np_recipe/run.yaml

@@ -1,12 +0,0 @@
-version: '1.0'
-steps:
-- graph_optimize: null
-- graph_quantize:
-  - scheme: W8-A8_dynamic
-    automatic_mixed_precision: null
-profile:
-- OUTPUT_MODEL
-evaluate:
-- model: OUTPUT_MODEL
-  metric:
-  - ACCURACY