1
|
{"args": {"method": "ff", "dataset": "mnist", "depth": 3, "width": 256, "epochs": 15, "batch_size": 128, "max_batches": 0, "eta": null, "momentum": 0.9, "act": "tanh", "residual": 0, "feedback_scale": 1.0, "pepita_f_scale": 0.05, "keep_prob": 0.9, "ff_threshold": 2.0, "ep_beta": 0.5, "ep_dt": 0.5, "ep_free_steps": 20, "ep_nudge_steps": 4, "canonical_preprocess": 1, "seed": 3, "device": "cuda", "outdir": "results", "tag": "baseline_budget_v1_mnist_ff_w256_d3_s3"}, "method_source": {"paper": "https://www.cs.toronto.edu/~hinton/absps/FFXfinal.pdf", "reference": "https://github.com/mpezeshki/pytorch_forward_forward", "protocol": "greedy layerwise goodness; input length normalization; Adam"}, "provenance": {"git_commit": "c8a372cae2612839bdbdafcf9c114f52cb4a6de3", "git_dirty": false}, "steps": [{"layer_end": 0, "test_acc": 0.1039, "local_loss": 0.24703803658485413}, {"layer_end": 1, "test_acc": 0.9392, "local_loss": 0.24402467906475067}, {"layer_end": 2, "test_acc": 0.9367, "local_loss": 0.2870331406593323}], "final": {"test_acc": 0.9367, "test_loss": NaN, "wall_s": 67.10431051254272}}
|
