Loading
Du hast einen ComfyUI-Workflow? Erstelle direkt im Browser eine verständliche Workflow-Dokumentation.Workflow verständlich machen.Jetzt ausprobieren ↗
This directory is the shared artifact location for the notebooks in ../notebooks/. The repository tracks the documentation and ignore rules here; model binaries are downloaded or generated locally when you run the...
Modellquelle
Quellenbeschreibung
This directory is the shared artifact location for the notebooks in
../notebooks/. The repository tracks the documentation and
ignore rules here; model binaries are downloaded or generated locally when you
run the workflows.
The same source model is used for three deployment paths:
Quellen
1 QuelleVerifiziert 16. Sept.
Modellartefakte
5 ArtefakteQuellenauszüge
2 Auszüge| Property | Value |
|---|---|
| Architecture | YOLOX-Small (LibreYOLOXs) |
| Framework | LibreYOLO |
| Task | Object detection |
| Base checkpoint | LibreYOLO/LibreYOLOXs |
| Input tensor | images |
| Input shape | 1 x 3 x 640 x 640 |
| Input format | BGR float32, NCHW, values in 0-255 |
LibreYOLOXs.pt
|
`---> LibreYOLOXs.onnx
|\
| `---> qairt/LibreYOLOXs_fp32.dlc
| `---> qairt/LibreYOLOXs_int8.dlc
|
| `---> snpe/LibreYOLOXs_fp32.dlc
| `---> snpe/LibreYOLOXs_int8.dlc
|
`---> qaihub/LibreYOLOXs_fp32.dlc
`---> qaihub/LibreYOLOXs_int8.dlc
All three notebooks reuse the same exported ONNX model.
| Path | Produced by | Description |
|---|---|---|
LibreYOLOXs.pt | All notebooks | Base PyTorch checkpoint, downloaded on demand |
LibreYOLOXs.onnx | All notebooks | Shared ONNX export used by all backend flows |
qairt/LibreYOLOXs_fp32.dlc | QAIRT notebook | FP32 DLC produced by qairt-converter |
qairt/LibreYOLOXs_int8.dlc | QAIRT notebook | INT8 DLC produced by qairt-quantizer |
qairt/binary/LibreYOLOXsInference.cpp | QAIRT notebook | C++ inference wrapper source |
qairt/binary/LibreYOLOXsInference.h | QAIRT notebook | C++ inference wrapper header |
qairt/binary/LibreYOLOXs_int8.bin | QAIRT notebook |
| Property | Value |
|---|---|
| Name | images |
| Shape | 1 x 3 x 640 x 640 |
| Layout | NCHW |
| Data type | float32 |
| Color order | BGR |
Preprocessing used by the notebooks:
640 x 640.114.0-255 range.The exported ONNX model is validated in each notebook before backend-specific compilation. The documentation in this directory assumes the shared decoded output layout used by the notebooks:
| Output | Shape | Description |
|---|---|---|
bboxes | 1 x 8400 x 4 | Bounding-box coordinates |
scores | 1 x 8400 x 81 | Objectness plus 80 COCO class scores |
INT8 flows use representative data prepared by the notebooks from COCO 2017 validation images:
docker/notebooks/dataset/docker/notebooks/subset/calib/docker/notebooks/subset/val/subset/calib/filenames.txt and subset/val/filenames.txtThe local QAIRT and SNPE quantizers read those manifests directly. The QAI Hub
notebook loads the same .raw tensors into NumPy arrays and submits them to the
cloud quantization job.
Published files are hosted on Hugging Face:
Example download:
huggingface-cli download fabricionarcizo/LibreYOLOXs \
--local-dir docker/models \
--exclude "*.gitattributes" "README.md"
You can also generate the same artifact layout locally by running the notebooks
in ../notebooks/.
The qairt/binary/ and qaihub/binary/ subdirectories contain a C++ inference
wrapper generated alongside each INT8 DLC. These files are created locally when
you run the respective notebook and are not Git-tracked.
qairt/binary/| File | Description |
|---|---|
LibreYOLOXsInference.cpp | C++ inference wrapper source |
LibreYOLOXsInference.h | C++ inference wrapper header |
LibreYOLOXs_int8.bin | Serialised INT8 model for the binary interface |
libLibreYOLOXs.so | Compiled shared library |
qaihub/binary/| File | Description |
|---|---|
LibreYOLOXsInference.cpp | C++ inference wrapper source |
LibreYOLOXsInference.h | C++ inference wrapper header |
LibreYOLOXs_int8.bin | Serialised INT8 model for the binary interface |
../README.md - Docker environment and setup../notebooks/README.md - notebook workflows and generated data layoutqaihub/binary/LibreYOLOXs_int8.bin
bin · 9,06 MB · SHA-256 83427220bc3a…571f · Hugging Face
qairt/binary/LibreYOLOXs_int8.bin
bin · 9,04 MB · SHA-256 a26e249ab686…12cd · Hugging Face
Herunterladen--- license: mit language: - en pipeline_tag: object-detection tags: - object-detection - yolox - onnx - dlc - qualcomm - snapdragon - qairt - snpe - qai-hub - edge-inference base_model: LibreYOLO/LibreYOLOXs datasets: - merve/coco --- # LibreYOLOXs Qualcomm-Optimized Model Artifacts This directory is the shared artifact location for the notebooks in [`../notebooks/`](../notebooks/). The repository tracks the documentation and ignore rules here; model binaries are downloaded or generated locally when you run the workflows. The same source model is used for three deployment paths: - **QAIRT** local conversion and quantization - **SNPE** local conversion and quantization - **QAI Hub** cloud compilation, quantization, and profiling --- ## Model overview | Property | Value | |---|---| | Architecture | YOLOX-Small (LibreYOLOXs) | | Framework | [LibreYOLO](https://github.com/LibreYOLO/libreyolo) | | Task | Object detection | | Base checkpoint | [LibreYOLO/LibreYOLOXs](https://huggingface.co/LibreYOLO/LibreYOLOXs) | | Input tensor | `images` | | Input shape | `1 x 3 x 640 x 640` | | Input format | BGR `float32`, NCHW, values in `0-255` | --- ## Artifact flow ```text LibreYOLOXs.pt | `---> LibreYOLOXs.onnx |\ | `---> qairt/LibreYOLOXs_fp32.dlc | `---> qairt/LibreYOLOXs_int8.dlc | | `---> snpe/LibreYOLOXs_fp32.dlc | `---> snpe/LibreYOLOXs_int8.dlc | `---> qaihub/LibreYOLOXs_fp32.dlc `---> qaihub/LibreYOLOXs_int8.dlc ``` All three notebooks reuse the same exported ONNX model. --- ## Expected files | Path | Produced by | Description | |---|---|---| | `LibreYOLOXs.pt` | All notebooks | Base PyTorch checkpoint, downloaded on demand | | `LibreYOLOXs.onnx` | All notebooks | Shared ONNX export used by all backend flows | | `qairt/LibreYOLOXs_fp32.dlc` | QAIRT notebook | FP32 DLC produced by `qairt-converter` | | `qairt/LibreYOLOXs_int8.dlc` | QAIRT notebook | INT8 DLC produced by `qairt-quantizer` | | `qairt/binary/LibreYOLOXsInference.cpp` | QAIRT notebook | C++ inference wrapper source | | `qairt/binary/LibreYOLOXsInference.h` | QAIRT notebook | C++ inference wrapper header | | `qairt/binary/LibreYOLOXs_int8.bin` | QAIRT notebook | Serialised INT8 model for the binary interface | | `qairt/binary/libLibreYOLOXs.so` | QAIRT notebook | Compiled shared library | | `snpe/LibreYOLOXs_fp32.dlc` | SNPE notebook | FP32 DLC produced by `snpe-...
Source context: 0 downloads · 1 likes · Pipeline object-detection · Repo fabricionarcizo/LibreYOLOXs
| Serialised INT8 model for the binary interface |
qairt/binary/libLibreYOLOXs.so | QAIRT notebook | Compiled shared library |
snpe/LibreYOLOXs_fp32.dlc | SNPE notebook | FP32 DLC produced by snpe-onnx-to-dlc |
snpe/LibreYOLOXs_int8.dlc | SNPE notebook | INT8 DLC produced by snpe-dlc-quantize |
qaihub/LibreYOLOXs_fp32.dlc | QAI Hub notebook | FP32 DLC downloaded from a cloud compile job |
qaihub/LibreYOLOXs_int8.dlc | QAI Hub notebook | INT8 DLC downloaded from cloud quantize + compile jobs |
qaihub/binary/LibreYOLOXsInference.cpp | QAI Hub notebook | C++ inference wrapper source |
qaihub/binary/LibreYOLOXsInference.h | QAI Hub notebook | C++ inference wrapper header |
qaihub/binary/LibreYOLOXs_int8.bin | QAI Hub notebook | Serialised INT8 model for the binary interface |