Furiosa-models

New Features

Improvements
- Added YOLOv5 Python postprocess 185
- Added parameter in YOLO postprocess to perform sigmoid or not 185

Removed

Bug fixes
- Add missing `requests` dependency 187

New Features
- Added YOLOv7w6 Pose Estimation Model 173
- Added `serve` subcommand to open REST server 182

Improvements
- Detailed explanation about `with_scaling` argument 171
- Added Jupyter Notebook examples 167
- Use new furiosa-compiler CLI 174
- Use IR version instead of compiler version when probing ENF files 174
- Unify postprocess signatures 180
- Use Typer CLI library instead of argparse 181

Removed
- furiosa-libcompiler APT package dependency 174

Bug fixes

New Features
- Provide 1pe artifacts too 158

Improvements
- Resolve artifact binaries lazily 155
- Added `ruff` linter 160
- Automatically build documentation on comment 162
- Upgrade `pydantic` library version to `2.0.0` 166
- Added `model.resolve_all()` to resolve all lazily loaded fields at once 166
- Added local artifact binary cache 166

Removed
- Removed unused `timm` dependency 149
- **Breaking:** Now uses default Python initializer instead of `model.load()` 166
- **Breaking:** `model.enf` field has been removed 166
- **Breaking:** `model.source`, `model.calib_yaml` fields have been renamed 166

New Features
- Add EfficientNetB0 model 121
- Add EfficientNetV2-S model 130
- Set default target as Warboy's production revision (B0) 125
- Provide calibration ranges for every model 144

Improvements
- Removed `Quantize` external operators 144
- Detailed error messages for model file fetching 144
- ENF generator can do the jobs parallelly 144
- Removed furiosa.registry dependency 144
- Faster import for furiosa.models 117
- Replace yolov5's box decode implementation in Rust 109
- Remove Cpp postprocessor implementations 102
- Change packaging tool from setuptools-rust to flit 109

Removed
- Truncated models and corresponding postprocesses 144
- **Breaking:** drop support of directly passing Model to session.create() 144

Tasks
- Release guide for developers 129
- Report regression test's result with PR comment 110

Bug Fixes
- Fix CLI to properly report net inference time 112
- Update certain model sources with valid onnx 120

New Features
- Add ResNet50 model
- Add SSD ResNet34 model
- Add SSD MobileNet model
- Add YOLOv5l model
- Add YOLOv5m model

Improvements
- Add native postprocessing implementation for ResNet50 42
- Add native postprocessing implementation for SSD ResNet34 45
- Add native postprocessing implementation for SSD MobileNet 16
- Refactor Model API to use classmethods to load models 66
- Make Model Zoo's ABC Model 78
- Retain only necessary python dependencies 81
- Improve enf_generator.sh and add enf, dfg models for e2e tests 33
- Add mkdocstrings to make API references 22

Tasks
- Regresstion Test 61
- Attach Tekton CI 41
- Yolov5 L/M e2e testing code 22
- Change the documentation layout and add resnet34, resnet50, mobilenet details 33
- Replace maturin with setuptools-rust 26

Bug Fixes
- Resolve DVC directly 80