GPU Detection Fix

When running training with Elephant on the dataset, the process may default to the CPU. To ensure the training runs on the GPU, the following changes were made to the Makefile in the Docker/elephant-server directory.

1. Force GPU usage in the launch target

In Docker/elephant-server/Makefile, within the launch target:

Before :

$(ELEPHANT_DOCKER) run -it --rm $(GPU_ARG) --shm-size=8g -v $(ELEPHANT_WORKSPACE):/workspace \

After :

$(ELEPHANT_DOCKER) run -it --rm $(GPU_ARG) --gpus all --shm-size=8g -v $(ELEPHANT_WORKSPACE):/workspace \

This change ensures that the container explicitly requests access to all GPUs.

2. Improve GPU detection logic in the warmup target

In Docker/elephant-server/Makefile, the warmup rule was updated to define a fallback behavior when ELEPHANT_GPU is not set.

Before :

warmup:
	$(eval GPU_ARG:=$(shell \
	if [ -n "$(ELEPHANT_NVIDIA_GID)" ] && [ -n "$(ELEPHANT_GPU)" ]; then \
		VAR=$$(echo --gpus '"device=$(ELEPHANT_GPU)"'); \
	fi;\
	echo $$VAR))
	@if [ -n "$(GPU_ARG)" ]; then \
		$(ELEPHANT_DOCKER) run -it --rm $(GPU_ARG) $(ELEPHANT_IMAGE_NAME) echo "warming up GPU..."; \
	else \
		echo "CPU mode..."; \
	fi

After :

warmup:
	$(eval GPU_ARG:=$(shell \
	if [ -n "$(ELEPHANT_NVIDIA_GID)" ] && [ -n "$(ELEPHANT_GPU)" ]; then \
		VAR="--gpus device=$(ELEPHANT_GPU)"; \
	else \
		VAR="--gpus all"; \
	fi; \
	echo $$VAR))

This ensures that when a specific GPU is not defined, the Docker container still uses all available GPUs instead of falling back to CPU execution.

Contributors: Jan Amouroux, Célia Brahimi, and Camille-Astrid Rodrigues
Master's students in Bioinformatics and Systems Biology at the University of Toulouse
contributed to this pull request as part of their supervised research project within the DEMO team (MCD laboratory, CBI-Toulouse).