SPIRA v2: entropy-patch tokenization + hierarchical dopamine-gated spiking cortex + hippocampal one-shot memory with replay consolidation
032a961 verified | #!/usr/bin/env python3 | |
| """Decisive roundtrip test: fresh 60k-tick train -> measure -> save -> load -> measure.""" | |
| import sys | |
| import numpy as np | |
| sys.path.insert(0, "/home/z/my-project/download/spira") | |
| from spira import SPIRA, SPIRAConfig | |
| from spira.hashing import fnv1a64 | |
| data = open("/home/z/my-project/data/train.bin", "rb").read() | |
| cfg = SPIRAConfig(snn_tokens=60_000, am_tokens=2000, epochs=1) | |
| model = SPIRA.build(data, cfg, log=lambda m: None) | |
| ids = model._ids_train | |
| N = 60_000 | |
| neurons = SPIRA.input_neurons_stream(ids[:N], cfg.d_in, cfg.k_in) | |
| cols = model.cols_of(ids[: N + 1]) | |
| cov = (cols[1:] != cfg.vocab_size)[: N - 1].mean() | |
| res = model.core.train_pass(neurons[: N - 1], cols[1:N], log_every=30_000, log_fn=lambda d: None) | |
| print(f"IN-PROCESS training: overall={res['accuracy']:.4f} in-vocab~={res['accuracy']/cov:.4f}") | |
| # in-process eval on the SAME slice (no learning) | |
| ev = model.eval_stream(data[:260_000], max_positions=12_000) | |
| print(f"IN-PROCESS eval (same slice): snn_invocab={ev['snn_acc_invocab']:.4f} " | |
| f"cov={ev['vocab_coverage']:.3f}") | |
| model.save("/tmp/roundtrip.npz") | |
| model2 = SPIRA.load("/tmp/roundtrip.npz") | |
| ev2 = model2.eval_stream(data[:260_000], max_positions=12_000) | |
| print(f"POST-LOAD eval (same slice): snn_invocab={ev2['snn_acc_invocab']:.4f}") | |
| # compare weight matrices directly | |
| same = np.allclose(model.core.W_out, model2.core.W_out, atol=1e-6) | |
| print(f"W_out identical after roundtrip: {same}") | |
| print(f"W_out col norm stats: max={np.linalg.norm(model.core.W_out, axis=0).max():.4f} " | |
| f"mean={np.linalg.norm(model.core.W_out, axis=0).mean():.4f}") | |
| print(f"threshold range: [{model2.core.thr.min():.3f}, {model2.core.thr.max():.3f}] " | |
| f"mean={model2.core.thr.mean():.3f}") | |
| print(f"x_mean norm: {np.linalg.norm(model2.core.x_mean):.3f} vs in-proc {np.linalg.norm(model.core.x_mean):.3f}") | |