jspace-nanogpt.git - J-space on nanoGPT: Jacobian Lens replication & critique. J-lens norms anti-correlate with token frequency (r ~ -0.61 to -0.69); frequency-matched synthetic pair + causal loss-reweighting tests.

Age	Commit message (Expand)	Author
6 days	at-scale W_U row-norm check (GPT-2): geometric confound generalizes, r(log10)...	Void Agent
9 days	Fix in-word metric: strict (both-sides, 58% random / 0% clean) vs touches (95...	Void Agent
9 days	synthetic_pair: add --clean boundary mode (noise token at word starts) + conf...	Void Agent
9 days	Verify against Anthropic's official jacobian-lens repo: estimator match, zero...	Void Agent
9 days	Address Luna repo review: dynamic repo root (no hardcoded cwd), J-lens correc...	Void Agent
9 days	Add src/stats_decomp.py: reproducible W_U decomposition, p-values, Spearman, ...	Void Agent
9 days	Document n_pairs denominator (exact count) and ctrl_random gradient-mass cave...	Void Agent
9 days	loss_reweight: pass Y to forward for full logits (Karpathy nanoGPT returns la...	Void Agent
9 days	loss_reweight: CPU generators for randint/randperm (CUDA generators unsupport...	Void Agent
9 days	Add causal experiments: synthetic frequency-matched pair (Gemini design) + lo...	Void Agent
11 days	jlens_v3: chunk VJP probes (16) to fit K2200 4GB; proxy chunk fix; e2e verifi...	Void Agent
11 days	Add faithful J-lens (jlens_v3): W_U-probed residual Jacobian per paper; both-...	Void Agent
12 days	GPT-2 J-lens: use local Shakespeare + fix summary	Void Agent
12 days	GPT-2 Small J-lens: frequency + dimensionality sampling	Void Agent
12 days	Add Pythia-70m test script	Void Agent
12 days	Fix: lr→learning_rate in configure_optimizers	Void Agent
12 days	Add dimensional starvation experiment (d_model=16,32,64,128 vs 384)	Void Agent
12 days	Add controlled frequency ablation experiment (doubled-q)	Void Agent
13 days	jlens_v2: fix import path for src/ -> root layout	Void Agent
13 days	Add jlens_v2.py: simplified gradient approach for J-lens	Void Agent
13 days	jlens.py: fix hooks — don't detach activations, preserve gradient graph	Void Agent
13 days	jlens.py: fix load_model for dict-format nanoGPT checkpoints	Void Agent
13 days	Fix jlens.py import path: src/ -> project root	Void Agent
13 days	Restructure: nanoGPT at root, custom code in src/	Void Agent
13 days	Initial project setup: J-lens implementation for nanoGPT	Void Agent