docs: refine rock5b ai documentation

[Ronin-1124]

Description of changes

docs: refine rock5b ai documentation

[Ronin-1124]

/windsurf-review

[windsurf-bot]

Other comments (18)

• docs/common/orion-common/app-dev/artificial-intelligence/_minicpm-v-2-6.mdx (123-123) The model testing command references LLaVA model paths (`./llava-v1.6-vicuna-7b/llava-v1.6-vicuna-7B-Q5_K_M.gguf`) but this documentation is for MiniCPM-V 2.6. The command should use the MiniCPM-V paths that were created in the previous steps.

  ./build/bin/llama-mtmd-cli -m ./MiniCPM-V-2_6/model/ggml-model-Q5_K_M.gguf --mmproj ./MiniCPM-V-2_6/mmproj-model-f16.gguf -p "What is this picture about?" --image ./tools/mtmd/test-1.jpeg
  

• docs/common/orion-common/app-dev/artificial-intelligence/_minicpm-v-2-6.mdx (6-6) The documentation states the model has "参数量约为 80 亿" (approximately 80 billion parameters), but the model output clearly shows "model params = 7.61 B". This should be corrected to avoid confusion about the model size.
• docs/common/ai/rockchip/_mobilesam.mdx (63-66) There appears to be an error in the model conversion commands. The encoder conversion is using the decoder model file path.

  cd ../python/decoder/
  python convert.py ../../model/mobilesam_decoder.onnx ${TARGET_PLATFORM}
  cd ../encoder/
  python convert.py ../../model/mobilesam_encoder.onnx ${TARGET_PLATFORM}
  

• docs/common/orion-common/app-dev/artificial-intelligence/_mobilenet-v2-int8.mdx (35-35) There's an inconsistency in the script naming between sections. The quick start section uses `inference_npu.py` (line 35) while the complete conversion process section uses `inference_onnx.py` (lines 81, 104, 114). These should be consistent to avoid confusion for users following the documentation.
• docs/common/orion-common/app-dev/artificial-intelligence/_llava-1-6-7b.mdx (141-141) The model path in the test command doesn't match the actual model being used. The command refers to `./MiniCPM-V-2_6/model/ggml-model-Q5_K_M.gguf` but should use the LLaVA model path instead.

  ./build/bin/llama-mtmd-cli -m ./llava-v1.6-vicuna-7b/llava-v1.6-vicuna-7B-Q5_K_M.gguf --mmproj ./llava-v1.6-vicuna-7b/vit/mmproj-model-f16.gguf -p "What is this picture about?" --image ./tools/mtmd/test-1.jpeg
  

• docs/common/ai/rockchip/_lprnet.mdx (154-154) The command uses `${TARGET_PLATFORM}` variable, but there's no instruction on how to set this variable in the device environment. Consider adding instructions to set this variable before running the Python example.
• docs/common/ai/rockchip/_yolov8.mdx (88-88) The SCP command uses placeholder values (`user@your_device_ip:target_directory`) that need to be replaced with actual values. Consider adding a note to make this clear to users.
• docs/common/ai/rockchip/_lprnet.mdx (88-88) The SCP command uses placeholder values (`user@your_device_ip:target_directory`) that need to be replaced with actual values. Consider adding a note to remind users to replace these with their specific device IP address and target directory path.
• docs/common/ai/rockchip/_yolov8.mdx (170-170) The command uses `${TARGET_PLATFORM}` variable which may not be defined in the device environment. Consider either adding instructions to define this variable on the device or update the example to use a specific platform like in the output example (`--target rk3588`).
• docs/common/ai/rockchip/list.mdx (1-13) I notice this file contains model descriptions in Chinese. If the rest of the documentation is in English, consider translating these descriptions for consistency. For example:

  yolov8 (Object Detection + Extended Features)
  ppyoloe (PaddleDetection Representative)
  yolo_world (Open Vocabulary Detection)
  yolov8_seg (Segmentation)
  mobilesam (Lightweight SAM)
  deeplabv3 (Classic Semantic Segmentation)
  PPOCR (Comprehensive OCR Solution)
  LPRNet (License Plate Recognition)
  RetinaFace (Face Detection)
  resnet (Basic Network)
  mobilenet (Lightweight Network)
  clip (Multimodal)
  whisper (Speech Recognition)
  

• docs/common/ai/rockchip/_wav2vec2.mdx (152-152) In the Python API example, the command uses `${TARGET_PLATFORM}` variable, but this variable is not set in the device environment. Users should either set this variable first or replace it with the actual target platform (e.g., 'rk3588' or 'rk356x').

  python wav2vec2.py --model_path ../model/wav2vec2_base_960h_20s.rknn --target rk3588  # or your specific target platform
  

• docs/common/ai/rockchip/_wav2vec2.mdx (88-88) The file synchronization command contains placeholders (`user@your_device_ip:target_directory`) that need to be replaced with actual values. Consider adding a note to make this clear to users.
• docs/common/ai/rockchip/_yolo_world.mdx (234-234) The example output shows an error message `write image fail! ret=1 path=out.png` but the documentation doesn't address this issue. Consider adding a note explaining this error (e.g., possible permission issues or required directory setup) or updating the example to show successful image writing.
• docs/common/orion-common/app-dev/artificial-intelligence/_qwen2-5-vl-3b.mdx (114-114) There's an inconsistency in the path to the test image. The command uses `./test.png` but the example output shows it was run with `./Qwen2.5-VL-3B-Instruct/test.png`. Consider updating either the command or ensuring the test image is available in both locations to avoid confusion.
• docs/common/ai/rockchip/_mobilenet.mdx (56-58) There's a grammatical error in this tip section.

  :::tip
  本示例可以在运行转换脚本之后看到模型输出。
  :::
  

• docs/common/ai/rockchip/_resnet.mdx (57-57) There's a grammatical error in this tip. The current text says "本示例在可以在运行转换脚本之后看到模型输出" which has a redundant "在". It should be "本示例可以在运行转换脚本之后看到模型输出" (removing the first "在").
• docs/common/dev/_rknn-install.mdx (129-129) There's an inconsistency in the code block type. Earlier blocks use `type="PC"` but this one uses `type="host"`. For consistency, consider using the same type parameter throughout the document.

  <NewCodeBlock tip="X86 Linux PC" type="PC">
  

• docs/common/orion-common/app-dev/artificial-intelligence/_mobilenet-v2-int8.mdx (104-114) There's an inconsistency in the case of the `--EP` parameter. It's shown as uppercase `NPU` in the instructions (lines 35, 104) but the example output shows it was run with lowercase `npu` (line 114). This should be consistent to avoid confusion.

💡 To request another review, post a new comment with "/windsurf-review".

[windsurf-bot]

Other comments (11)

• docs/common/ai/rockchip/_mobilesam.mdx (63-66) There's an error in the model conversion commands. The encoder conversion is incorrectly using the decoder model file path.

  cd ../python/decoder/
  python convert.py ../../model/mobilesam_decoder.onnx ${TARGET_PLATFORM}
  cd ../encoder/
  python convert.py ../../model/mobilesam_encoder.onnx ${TARGET_PLATFORM}
  

• docs/common/ai/rockchip/_mobilenet.mdx (115-117) The directory name in this instruction doesn't match the directory name used in the file sync step (rknn_mobilenet_demo).

  cd rknn_mobilenet_demo/
  export LD_LIBRARY_PATH=./lib
  

• docs/common/ai/rockchip/_yolov8_seg.mdx (100-100) There's a directory path mismatch in the C API section. The instruction says to change to `rknn_yolov8_demo/` but the actual directory name appears to be `rknn_yolov8_seg_demo/` based on the file sync instructions in the previous section.
• docs/common/ai/rockchip/_yolov8_seg.mdx (180-180) The Python example uses `${TARGET_PLATFORM}` variable, but there's no instruction to set this variable in the Device environment. Users should be instructed to set this variable (e.g., `export TARGET_PLATFORM=rk3588`) before running the Python example.
• docs/common/ai/rockchip/_lprnet.mdx (88-88) The SCP command uses placeholder values (`user@your_device_ip:target_directory`) that need to be replaced with actual values. Consider adding a note to remind users to replace these with their specific device IP address and target directory path.
• docs/common/ai/rockchip/_lprnet.mdx (154-154) The command uses `${TARGET_PLATFORM}` variable, but there's no instruction to set this variable in the device environment. Consider adding instructions to set this variable before running the Python example, similar to how it's done in the model conversion section.
• docs/common/ai/rockchip/list.mdx (1-13) I notice this file contains model descriptions in Chinese. If the documentation is primarily in English, consider translating these descriptions for better international accessibility. For example:

  yolov8 (Object Detection + Extended Features)
  ppyoloe (PaddleDetection Representative)
  yolo_world (Open Vocabulary Detection)
  yolov8_seg (Segmentation)
  mobilesam (Lightweight SAM)
  deeplabv3 (Classic Semantic Segmentation)
  PPOCR (Comprehensive OCR Solution)
  LPRNet (License Plate Recognition)
  RetinaFace (Face Detection)
  resnet (Basic Network)
  mobilenet (Lightweight Network)
  clip (Multimodal)
  whisper (Speech Recognition)
  

• docs/common/ai/rockchip/_mobilenet.mdx (56-58) There's a grammatical error in this tip section.

  :::tip
  本示例可以在运行转换脚本之后看到模型输出。
  :::
  

• docs/common/ai/rockchip/_resnet.mdx (57-57) There's a grammatical error in this tip. The sentence structure is incorrect.

  :::tip
  本示例可以在运行转换脚本之后看到模型输出。
  :::
  

• docs/common/orion-common/app-dev/artificial-intelligence/_mobilenet-v2-int8.mdx (35-35) There's an inconsistency in the script naming. The quick start section uses `inference_npu.py` (line 35) while the complete conversion process section uses `inference_onnx.py` (lines 81, 104, 114). These should be consistent to avoid confusion for users following the documentation.
• docs/common/orion-common/app-dev/artificial-intelligence/_mobilenet-v2-int8.mdx (104-114) There's a case inconsistency with the `--EP` parameter. The documentation shows uppercase `NPU` in the command examples (lines 35, 104), but the example output shows lowercase `npu` (line 114). This should be consistent to avoid confusion.

💡 To request another review, post a new comment with "/windsurf-review".

[Ronin-1124]

/windsurf-review

[windsurf-bot]

Other comments (8)

• docs/common/ai/rockchip/_lprnet.mdx (88-88) The SCP command uses placeholder values (`user@your_device_ip:target_directory`) that need to be replaced with actual values. Consider adding a note to remind users to replace these with their specific device IP address and target directory path.
• docs/common/ai/rockchip/_clip.mdx (153-153) The test image is referenced as 'resnet-test.webp' which suggests it's related to ResNet rather than CLIP. Consider renaming this image to something CLIP-specific (like 'clip-test.webp' or 'dog-clip-test.webp') to maintain consistency with the example being demonstrated.
• docs/common/ai/rockchip/_resnet.mdx (103-103) The SCP command uses generic placeholders (`user@your_device_ip:target_directory`). Consider providing more concrete examples or instructions for users to determine their device IP and appropriate target directory, such as:

  scp -r rknn_resnet_demo/ rock@192.168.1.X:/home/rock/
  # Replace 192.168.1.X with your device's actual IP address
  

• docs/common/ai/rockchip/_mobilenet.mdx (103-103) The SCP command uses placeholders (`user@your_device_ip:target_directory`) which might be confusing for users. Consider providing more specific guidance, such as:

  scp -r rknn_mobilenet_demo/ rock@192.168.1.X:/home/rock/
  

  With a note explaining that users should replace the IP address with their device's actual IP.

• docs/common/orion-common/app-dev/artificial-intelligence/_qwen2-5-vl-3b.mdx (114-114) There's an inconsistency in the path to the test image. In the quick start section, the path is './test.png', but in the model testing section output, it shows './Qwen2.5-VL-3B-Instruct/test.png'. Consider updating the documentation to use consistent paths throughout.
• docs/common/ai/rockchip/_clip.mdx (192-192) There's an inconsistency in the example outputs. In the C API example (line 148), the text input is shown as 'a photo of a dog', but in the Python API example (line 192), it's shown as 'a photo of dog' (missing the article 'a'). Consider making these consistent to avoid confusion.
• docs/common/orion-common/app-dev/artificial-intelligence/_mobilenet-v2-int8.mdx (114-114) There's an inconsistency in the case of the 'NPU' parameter. In the command example it's uppercase (`--EP NPU`), but in the output example it's lowercase (`--EP npu`). For clarity, it would be good to make these consistent, especially if the parameter is case-sensitive.
• docs/common/ai/rockchip/_yolov8_seg.mdx (111-111) The executable name should match the directory name mentioned earlier for consistency.

  ./rknn_yolov8_seg_demo ./model/yolov8_seg.rknn ./model/bus.jpg
  

💡 To request another review, post a new comment with "/windsurf-review".

[Ronin-1124]

All done, ready to merge.