PRIMA / code /docs /installation.md

Remove debug instrumentation and rename the M4Seg dataset adapter

5e66a57 verified about 17 hours ago

1 kB

# Installation

The initial implementation targets Python 3.10, PyTorch 1.13.1 with CUDA 11.7, and Transformers 4.42.3. Inference uses BF16. This environment has been used for development; a fresh installation remains to be validated.

```bash
conda create -n prima python=3.10 -y
conda activate prima
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 \
  --extra-index-url https://download.pytorch.org/whl/cu117
pip install -r requirements.txt
```

COCO caption metrics require a Java runtime. Training writes progress and scalar losses to standard output.

MMCV and MMDetection are not required. Run commands from the project root. Placeholders must be replaced before loading a model or dataset.

Run the release interface checks with:

```bash
python -m unittest discover -s tests -v
```

These tests cover checkpoint loading, grounding-mask alignment, and evaluation behavior without downloading model weights or datasets. Test caches and generated outputs are excluded by `.gitignore`.

	# Installation

	The initial implementation targets Python 3.10, PyTorch 1.13.1 with CUDA 11.7, and Transformers 4.42.3. Inference uses BF16. This environment has been used for development; a fresh installation remains to be validated.

	```bash
	conda create -n prima python=3.10 -y
	conda activate prima
	pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 \
	--extra-index-url https://download.pytorch.org/whl/cu117
	pip install -r requirements.txt
	```

	COCO caption metrics require a Java runtime. Training writes progress and scalar losses to standard output.

	MMCV and MMDetection are not required. Run commands from the project root. Placeholders must be replaced before loading a model or dataset.

	Run the release interface checks with:

	```bash
	python -m unittest discover -s tests -v
	```

	These tests cover checkpoint loading, grounding-mask alignment, and evaluation behavior without downloading model weights or datasets. Test caches and generated outputs are excluded by `.gitignore`.