Artem Plastinkin commited on
Commit
8697afc
·
1 Parent(s): 7b5d384

Remove duplicate benchmark

Browse files
Files changed (1) hide show
  1. int8/benchmarks/x5h_mwmx_npu.yaml +0 -69
int8/benchmarks/x5h_mwmx_npu.yaml DELETED
@@ -1,69 +0,0 @@
1
- hardware:
2
- vendor: renesas
3
- chip: rcar-x5h
4
- cpu: arm-cortex-a720
5
- npu: npx6-48k
6
- npu_count: 2
7
- npu_cores: 12
8
- npu_default_freq_mhz: 1066
9
- accelerator:
10
- - npu
11
-
12
- runtime:
13
- engine: mwmx # Renesas MWMX (Middleware MX) native inference runtime
14
- toolchain_version: "MWMX SDK v4.35.0"
15
- format: onnx # input model format (FP32 ONNX)
16
- execution_provider: npu
17
- execution_precision: int8 # FP32 model is auto-cast to INT8 by the MWMX toolchain
18
-
19
- configuration:
20
- npu_instances: 1 # single NPU
21
- npu_cores_per_instance: 12 # single AI core
22
- npu_freq_mhz: 850 # 850 MHz NPU clock frequency
23
-
24
- benchmark:
25
- type: hil # Hardware-in-the-loop — measured on physical X5H silicon
26
- parameters:
27
- batch_size: 1
28
- input_resolution: [1, 3, 480, 640]
29
-
30
- performance:
31
- fps: null # throughput: 1000 / latency
32
- latency: 7.30 # median inference latency (ms) over 1000 warm runs
33
-
34
- metrics:
35
- accuracy: null
36
- top5_accuracy: null
37
-
38
- memory:
39
- peak_mb: null # peak NPU memory during inference
40
-
41
- power:
42
- avg_w: null # not yet characterized
43
-
44
- # Same config (1 instance, 12 cores, 850 MHz) as x5h_mwmx_npu_apm80_12core.yaml
45
- # — reusing its verified reproduce steps rather than re-deriving them.
46
- reproduce:
47
- steps:
48
- - title: Activate the Python environment
49
- command: >-
50
- Activate the Python virtual environment that has the `hf` CLI
51
- (huggingface_hub) and the NNAC toolchain installed, e.g.
52
- `source nnac_venv/bin/activate` -- path depends on your toolchain install.
53
- kind: note
54
- - title: Download the ONNX model and compile config
55
- command: hf download Renesas/RetinaNet-R101-ONNX --repo-type model --include "fp32/*" "compile_config/*" --local-dir ./RetinaNet-R101-ONNX-fp32
56
- - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
57
- command: |
58
- python3 nnac_frontend/legalize.py -d binary/nnx ./RetinaNet-R101-ONNX-fp32/fp32/retinanet-9.onnx --num-core 12 --network-config ./RetinaNet-R101-ONNX-fp32/compile_config/network_config.yaml
59
- - title: Set up the R-Car X5H board
60
- 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.
61
- kind: note
62
- - title: Copy the compiled artifact to the board
63
- 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.).
64
- kind: note
65
- - title: Run on R-Car X5H (single NPU cluster, 12 AI cores)
66
- command: |
67
- cd binary
68
- ./host_app ./arc_prog_npus ./nnx/retinanet-9
69
- 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.