Update performance numbers

int8/benchmarks/x5h_mwmx_npu_apm80_12core.yaml

@@ -29,7 +29,7 @@ benchmark:
 
 performance:
   fps: null
-  latency: 6.810764        # APM80 pipeline run (metawaremx_runtime CI)
+  latency: 6.72               # synced to mwmx2.2 model_list.html (2026-09-22); was 6.810764 (APM80 pipeline run, metawaremx_runtime CI)
 
 metrics:
   accuracy: null

int8/benchmarks/x5h_mwmx_npu_apm80_1core.yaml

@@ -0,0 +1,70 @@
+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             # Renesas MWMX (Middleware MX) native inference runtime
+  toolchain_version: "MWMX SDK v4.35.0"
+  format: onnx             # input model format (FP32 ONNX)
+  execution_provider: npu
+  execution_precision: int8  # FP32 model is auto-cast to INT8 by the MWMX toolchain
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 1
+  npu_freq_mhz: 850
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: [1, 3, 480, 640]
+
+performance:
+  fps: null
+  latency: 44.77               # synced to mwmx2.2 model_list.html (2026-09-22); no prior local measurement for this core count
+
+metrics:
+  accuracy: null
+  top5_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+# Exact commands verified against the NNAC "Getting Started" chapter. Rendered
+# by the AI-Dashboard in place of the generic placeholder flow — see
+# downloadRunHTML() / parse_reproduce() in AI-Dashboard/app.js.
+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/RetinaNet-R101-ONNX --repo-type=model --include "fp32/*" --local-dir ./RetinaNet-R101-ONNX-fp32
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./RetinaNet-R101-ONNX-fp32/fp32/retinanet-9.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/retinanet-9
+      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_apm80_3core.yaml

@@ -0,0 +1,72 @@
+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             # Renesas MWMX (Middleware MX) native inference runtime
+  toolchain_version: "MWMX SDK v4.35.0"
+  format: onnx             # input model format (FP32 ONNX)
+  execution_provider: npu
+  execution_precision: int8  # FP32 model is auto-cast to INT8 by the MWMX toolchain
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 3
+  npu_freq_mhz: 850
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: [1, 3, 480, 640]
+
+performance:
+  fps: null
+  latency: 15.38              # synced to mwmx2.2 model_list.html (2026-09-22); no prior local measurement for this core count
+
+metrics:
+  accuracy: null
+  top5_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+# UNCONFIRMED (segmentation config): this repo's segmentation_json compile_config
+# (compile_config/RetinaNet_l2seg.json, see the 12-core reproduce block) is
+# specific to the 12-core build; no 3-core segmentation config is bundled, so
+# the steps below compile without one — a best-effort single-cluster flow that
+# may not reproduce the exact benchmarked 3-core latency.
+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/RetinaNet-R101-ONNX --repo-type=model --include "fp32/*" --local-dir ./RetinaNet-R101-ONNX-fp32
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./RetinaNet-R101-ONNX-fp32/fp32/retinanet-9.onnx --num-core 3
+    - 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, 3 AI cores)
+      command: |
+        cd binary
+        ./host_app ./arc_prog_npus ./nnx/retinanet-9
+      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_apm80_4core.yaml

@@ -0,0 +1,72 @@
+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             # Renesas MWMX (Middleware MX) native inference runtime
+  toolchain_version: "MWMX SDK v4.35.0"
+  format: onnx             # input model format (FP32 ONNX)
+  execution_provider: npu
+  execution_precision: int8  # FP32 model is auto-cast to INT8 by the MWMX toolchain
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 4
+  npu_freq_mhz: 850
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: [1, 3, 480, 640]
+
+performance:
+  fps: null
+  latency: 11.66              # synced to mwmx2.2 model_list.html (2026-09-22); no prior local measurement for this core count
+
+metrics:
+  accuracy: null
+  top5_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+# UNCONFIRMED (segmentation config): this repo's segmentation_json compile_config
+# (compile_config/RetinaNet_l2seg.json, see the 12-core reproduce block) is
+# specific to the 12-core build; no 4-core segmentation config is bundled, so
+# the steps below compile without one — a best-effort single-cluster flow that
+# may not reproduce the exact benchmarked 4-core latency.
+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/RetinaNet-R101-ONNX --repo-type=model --include "fp32/*" --local-dir ./RetinaNet-R101-ONNX-fp32
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./RetinaNet-R101-ONNX-fp32/fp32/retinanet-9.onnx --num-core 4
+    - 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, 4 AI cores)
+      command: |
+        cd binary
+        ./host_app ./arc_prog_npus ./nnx/retinanet-9
+      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_apm80_6core.yaml

@@ -0,0 +1,72 @@
+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             # Renesas MWMX (Middleware MX) native inference runtime
+  toolchain_version: "MWMX SDK v4.35.0"
+  format: onnx             # input model format (FP32 ONNX)
+  execution_provider: npu
+  execution_precision: int8  # FP32 model is auto-cast to INT8 by the MWMX toolchain
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 6
+  npu_freq_mhz: 850
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: [1, 3, 480, 640]
+
+performance:
+  fps: null
+  latency: 9.24               # synced to mwmx2.2 model_list.html (2026-09-22); no prior local measurement for this core count
+
+metrics:
+  accuracy: null
+  top5_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+# UNCONFIRMED (segmentation config): this repo's segmentation_json compile_config
+# (compile_config/RetinaNet_l2seg.json, see the 12-core reproduce block) is
+# specific to the 12-core build; no 6-core segmentation config is bundled, so
+# the steps below compile without one — a best-effort single-cluster flow that
+# may not reproduce the exact benchmarked 6-core latency.
+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/RetinaNet-R101-ONNX --repo-type=model --include "fp32/*" --local-dir ./RetinaNet-R101-ONNX-fp32
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./RetinaNet-R101-ONNX-fp32/fp32/retinanet-9.onnx --num-core 6
+    - 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, 6 AI cores)
+      command: |
+        cd binary
+        ./host_app ./arc_prog_npus ./nnx/retinanet-9
+      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.