--cpu

🦉 AthenaAI is an AI-powered ensemble machine-learning trading bot for PocketOption, built on BinaryOptionsToolsV2 → Get it on GitLab
💬 Need help? Join the Chipa Discord · ⚡ Go no-code with ChipaEditor

Force the neural model to train on the CPU even if a CUDA or Apple MPS GPU is detected.

What it does

Trainer.__init__ calls detect_device(force_cpu=cfg.force_cpu) (src/training/device.py). With --cpu, detection short-circuits immediately and returns ("cpu", "CPU (forced via --cpu)") — torch is not even imported for detection. Without it, the order is: CUDA (torch.cuda.is_available(), reporting GPU name and VRAM), then Apple MPS, then CPU.

The chosen device is only used by _train_nn(), which passes it to TorchMLPClassifier(device=...); in fit() the network and each mini-batch are moved to that device. Everything else — CSV parsing, sample generation (multi-process CPU), sklearn models — is CPU-bound regardless of this flag.

Note the flag only affects training. At predict/load time, TorchMLPClassifier._ensure_net() re-detects the machine’s device on its own (cuda if available, else cpu).

Choosing a value

• Use it to keep the GPU free for other work, to rule out GPU driver/CUDA issues when debugging, or for reproducibility comparisons.
• CPU training is slower per epoch; the docstring example compensates with more epochs (--cpu --epochs 50), though epochs are independent of the device.
• If you don’t have a GPU there is no need for this flag — detection falls back to CPU automatically.

Examples

python train.py --cpu
python train.py --cpu --epochs 50

Interactions

• --check-gpu — shows what would be detected (without --cpu applied; --check-gpu calls detect_device() with no force flag).
• --no-nn — if the NN is skipped, the device choice only affects the startup log line.
• --batch-size — GPU memory pressure disappears on CPU, but large batches still cost RAM.

See also

• GPU guide
• src.training.device API reference

💬 Unsure how this interacts with the rest of the configuration? Ask in the Chipa Discord — or prototype strategies no-code with ChipaEditor.