Initial upload of EfficientNet-B2-ONNX

.metadata.yaml

@@ -0,0 +1,33 @@
+model:
+  name: efficientnet-b2
+  display_name: EfficientNet-B2
+  upstream: timm/tf_efficientnet_b2.ns_jft_in1k
+  source:
+    kind: openmmlab
+    id: efficientnet_b2_3rdparty_ra_noisystudent_in1k
+    url: https://github.com/open-mmlab/mmpretrain/blob/main/configs/efficientnet/metafile.yml
+    note: >-
+      weights ported from the TF EfficientNet release
+
+architecture:
+  family: efficientnet   # lineage only — no version, no size
+  version: "v1"
+  variant: b2
+  dataset: imagenet-1k
+  input_resolution: 260x260
+  num_classes: 1000
+  modality:
+    - vision
+  parameters: 9.2M
+  parameters_source: published
+  parameters_note: >-
+    published EfficientNet-B2 figure (Tan & Le 2019)
+
+format:
+  type: onnx
+  # opset: read it off the graph rather than guessing —
+  #   onnx.load('model.onnx').opset_import[0].version
+  # NOTE: `version` is a GGUF-only field and must not be used for ONNX.
+
+tasks:
+  - image-classification

README.md

@@ -0,0 +1,95 @@
+---
+license: apache-2.0
+base_model:
+- timm/tf_efficientnet_b2.ns_jft_in1k
+pipeline_tag: image-classification
+tags:
+- image-classification
+- computer-vision
+- renesas
+- x5h
+- onnx
+- efficientnet
+---
+
+# EfficientNet-B2 (ONNX) – Renesas X5H
+
+> ⏳ **Model file not yet uploaded.** Benchmark results on this page were published ahead of the model weights — see **Provided Artifacts** below. Download/deployment steps will not work until the file is added to this repository.
+
+## Introduction
+
+This repository hosts **EfficientNet-B2**, targeting the **Renesas R-Car X5H** platform for image
+classification inference on the NPX6 NPU.
+
+> **Note:** The other EfficientNet sizes (B0, B1, B3–B8) each have their own sibling repo.
+
+- **Model Architecture:** EfficientNet — compound-scaled convolutional network for 1000-class image classification
+- **Source Model:** [timm/tf_efficientnet_b2.ns_jft_in1k](https://huggingface.co/timm/tf_efficientnet_b2.ns_jft_in1k) — OpenMMLab config [`efficientnet_b2_3rdparty_ra_noisystudent_in1k`](https://github.com/open-mmlab/mmpretrain/blob/main/configs/efficientnet/metafile.yml)
+- **Task:** Image Classification (ImageNet-1k, 1000 classes)
+- **Parameters:** 9.2M (published EfficientNet-B2 figure (Tan & Le 2019))
+
+## Deployment Flow
+
+The FP32 ONNX model is auto-cast to **INT8** by the Renesas MWMX toolchain at compile time — no
+separate quantization step is required.
+
+```
+efficientnet_b2_3rdparty_ra_noisystudent_in1k_..._optimized.onnx (FP32)
+        │
+        └─▶  MWMX Runtime  ──▶  INT8 auto-cast  ──▶  NPX6 NPU
+```
+
+## Provided Artifacts
+
+| Artifact | Status | Notes |
+|----------|--------|-------|
+| **FP32 (ONNX)** | ⏳ Not yet uploaded | Benchmark numbers below exist; the model file has not been published to this repo yet |
+
+## Performance
+
+Measured on **Renesas R-Car X5H** via the MWMX runtime (APM50 ship-performance CI pipeline).
+
+> **Benchmark configuration:** Single NPU · Batch size: 1 · Input resolution: not available from source data (TBD)
+
+| Runtime | Precision | Device | Latency (ms) | Type |
+|---------|-----------|--------|---------------|------|
+| MWMX Runtime | INT8 (auto) | X5H · 1× NPU · 1 Core · 850 MHz | 7.786218 | Measured |
+| MWMX Runtime | INT8 (auto) | X5H · 1× NPU · 12 Cores · 850 MHz | 6.624581 | Measured |
+
+### Accuracy
+
+TBD — not yet measured/published for this repo.
+
+---
+
+## Runtime Details
+
+### MWMX Runtime
+
+- **Engine:** Renesas MWMX (Middleware MX) native inference runtime
+- **Input format:** FP32 ONNX (compiled by the MWMX toolchain)
+- **NPU execution precision:** INT8 (auto-cast by MWMX toolchain)
+- **Execution target:** NPX6-48K NPU on R-Car X5H
+
+---
+
+## Prerequisites
+
+To run inference on Renesas R-Car X5H, you need:
+
+1. **Renesas R-Car X5H board** with NPX6 NPU
+2. **Renesas MWMX Runtime**
+3. **Hugging Face CLI** to download the model (once the model file is published)
+
+## Download
+
+TBD — model file not yet published to this repository.
+
+---
+
+## Benchmark Methodology
+
+- **HIL runs:** Hardware-in-the-loop — measured on physical R-Car X5H silicon via the MWMX
+  runtime (`metawaremx_runtime` CI pipeline, "APM50" ship-performance target)
+- **Precision:** FP32 ONNX input; INT8 execution (auto-cast by MWMX)
+- **Slices:** results reported for both 1 AI core and 12 AI cores per NPU instance

fp32/.metadata.yaml

@@ -0,0 +1,17 @@
+# EfficientNet-B2 — source checkpoint: efficientnet_b2_3rdparty_ra_noisystudent_in1k
+architecture:
+  parameters: 9.2M   # published EfficientNet-B2 parameter count (EfficientNet paper / timm)
+
+variant:
+  id: onnx_fp32
+  format: onnx
+  precision: fp32            # full-precision FP32 reference model
+  method: onnx_export        # exported from the upstream source checkpoint
+
+quantization:
+  datatype: fp32
+  scope:
+    - weights
+  granularity: none          # not applicable for FP32
+  calibration: none          # no calibration for FP32 reference
+  toolchain: onnx

fp32/efficientnet-b2_3rdparty-ra-noisystudent_in1k.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89c9f466a667b8b0da35f72e0b464924f8b4d59b8573ff77f1cab6837a243e41
+size 36677482

int8/.metadata.yaml

@@ -0,0 +1,18 @@
+# EfficientNet-B2 — source checkpoint: efficientnet_b2_3rdparty_ra_noisystudent_in1k
+architecture:
+  parameters: 9.2M   # published EfficientNet-B2 parameter count (EfficientNet paper / timm)
+
+variant:
+  id: onnx_int8
+  format: onnx
+  precision: int8            # INT8 quantized — auto-cast by NPU execution provider at runtime
+  method: onnx_export        # base model exported to ONNX; quantization applied by runtime
+
+quantization:
+  datatype: int8
+  scope:
+    - weights
+    - activations
+  granularity: per-tensor    # typical for runtime-cast INT8
+  calibration: ptq           # post-training quantization applied by the NPU EP / MWMX toolchain
+  toolchain: onnx

int8/benchmarks/x5h_mwmx_npu_apm50_12core.yaml

@@ -0,0 +1,67 @@
+hardware:
+  vendor: renesas
+  chip: rcar-x5h
+  cpu: arm-cortex-a720
+  npu: npx6-48k
+  npu_count: 2
+  npu_cores: 12
+  npu_default_freq_mhz: 1066
+  accelerator:
+    - npu
+
+runtime:
+  engine: mwmx
+  toolchain_version: "MWMX SDK v4.35.0"
+  format: onnx
+  execution_provider: npu
+  execution_precision: int8
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 12
+  npu_freq_mhz: 850
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: null   # not available from source data — TBD
+
+performance:
+  fps: null
+  latency: 6.624581          # APM50 pipeline run (metawaremx_runtime CI)
+
+metrics:
+  accuracy: null
+  top5_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+reproduce:
+  steps:
+    - title: Activate the Python environment
+      command: >-
+        Activate the Python virtual environment that has the `hf` CLI
+        (huggingface_hub) and the NNAC toolchain installed, e.g.
+        `source nnac_venv/bin/activate` -- path depends on your toolchain install.
+      kind: note
+    - title: Download the ONNX model
+      command: hf download Renesas/EfficientNet-B2-ONNX --repo-type=model --include "fp32/*" --local-dir ./EfficientNet-B2-ONNX-fp32
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./EfficientNet-B2-ONNX-fp32/fp32/efficientnet-b2_3rdparty-ra-noisystudent_in1k.onnx --num-core 12
+    - title: Set up the R-Car X5H board
+      command: Configure the board per the AI Compiler (NNAC) "Getting Started" guide, section 3.4 (host TFTP/NFS setup, bootloader flashing, U-Boot, Linux boot, login) -- exact steps depend on your board/network setup.
+      kind: note
+    - title: Copy the compiled artifact to the board
+      command: Copy ${WORKDIR}/binary/nnx (the working directory from the download/compile steps above) to the board -- method may vary (NFS mount, scp, USB, etc.).
+      kind: note
+    - title: Run on R-Car X5H (single NPU cluster, 12 AI cores)
+      command: |
+        cd binary
+        ./host_app ./arc_prog_npus ./nnx/efficientnet-b2_3rdparty-ra-noisystudent_in1k
+      expected: hash[n] = 0x...(OK) means the run's output matches the reference hash in hash.txt; latency is the NPX execution time reported in cycles and ms.

int8/benchmarks/x5h_mwmx_npu_apm50_1core.yaml

@@ -0,0 +1,67 @@
+hardware:
+  vendor: renesas
+  chip: rcar-x5h
+  cpu: arm-cortex-a720
+  npu: npx6-48k
+  npu_count: 2
+  npu_cores: 12
+  npu_default_freq_mhz: 1066
+  accelerator:
+    - npu
+
+runtime:
+  engine: mwmx
+  toolchain_version: "MWMX SDK v4.35.0"
+  format: onnx
+  execution_provider: npu
+  execution_precision: int8
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 1
+  npu_freq_mhz: 850
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: null   # not available from source data — TBD
+
+performance:
+  fps: null
+  latency: 7.786218          # APM50 pipeline run (metawaremx_runtime CI)
+
+metrics:
+  accuracy: null
+  top5_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+reproduce:
+  steps:
+    - title: Activate the Python environment
+      command: >-
+        Activate the Python virtual environment that has the `hf` CLI
+        (huggingface_hub) and the NNAC toolchain installed, e.g.
+        `source nnac_venv/bin/activate` -- path depends on your toolchain install.
+      kind: note
+    - title: Download the ONNX model
+      command: hf download Renesas/EfficientNet-B2-ONNX --repo-type=model --include "fp32/*" --local-dir ./EfficientNet-B2-ONNX-fp32
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./EfficientNet-B2-ONNX-fp32/fp32/efficientnet-b2_3rdparty-ra-noisystudent_in1k.onnx --num-core 1
+    - title: Set up the R-Car X5H board
+      command: Configure the board per the AI Compiler (NNAC) "Getting Started" guide, section 3.4 (host TFTP/NFS setup, bootloader flashing, U-Boot, Linux boot, login) -- exact steps depend on your board/network setup.
+      kind: note
+    - title: Copy the compiled artifact to the board
+      command: Copy ${WORKDIR}/binary/nnx (the working directory from the download/compile steps above) to the board -- method may vary (NFS mount, scp, USB, etc.).
+      kind: note
+    - title: Run on R-Car X5H (single NPU cluster)
+      command: |
+        cd binary
+        ./host_app ./arc_prog_npus ./nnx/efficientnet-b2_3rdparty-ra-noisystudent_in1k
+      expected: hash[n] = 0x...(OK) means the run's output matches the reference hash in hash.txt; latency is the NPX execution time reported in cycles and ms.