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 (Collapse)	Author
2026-07-31	jlens_v3: chunk VJP probes (16) to fit K2200 4GB; proxy chunk fix; e2e ↵	Void Agent
	verified (last-layer cos-sim=1.0)
2026-07-31	Add faithful J-lens (jlens_v3): W_U-probed residual Jacobian per paper; ↵	Void Agent
	both-ways comparison vs log-softmax proxy; 3-model adversarial reviews
2026-07-30	GPT-2 J-lens: use local Shakespeare + fix summary	Void Agent

2026-07-30	GPT-2 Small J-lens: frequency + dimensionality sampling	Void Agent

2026-07-30	Add Pythia-70m test script	Void Agent

2026-07-30	Fix: lr→learning_rate in configure_optimizers	Void Agent

2026-07-30	Add dimensional starvation experiment (d_model=16,32,64,128 vs 384)	Void Agent

2026-07-30	Add controlled frequency ablation experiment (doubled-q)	Void Agent

2026-07-29	jlens_v2: fix import path for src/ -> root layout	Void Agent

2026-07-29	Add jlens_v2.py: simplified gradient approach for J-lens	Void Agent

2026-07-29	jlens.py: fix hooks — don't detach activations, preserve gradient graph	Void Agent

2026-07-29	jlens.py: fix load_model for dict-format nanoGPT checkpoints	Void Agent

2026-07-29	Fix jlens.py import path: src/ -> project root	Void Agent

2026-07-29	Maxwell GPU fixes: disable bf16 SDPA, force fp32	Void Agent
	- model.py: honor config.flash flag (defaults True, False disables SDPA) - config: force dtype=float32 and flash=False for K2200 (compute 5.0) - Maxwell GPUs don't support bf16; SDPA internally uses bf16 operations
2026-07-29	Restructure: nanoGPT at root, custom code in src/	Void Agent
	- Move model.py, train.py, configurator.py to root for nanoGPT compatibility - data/ and config/ directories at root with Shakespeare dataset prep scripts - src/jlens.py updated to import model from project root - Cleaned up stale src/config/ and duplicate src/ files - Fixed .gitignore: exclude out-shakespeare-char/ instead of raw data dirs
2026-07-29	Initial project setup: J-lens implementation for nanoGPT	Void Agent
	- Core J-lens computation (batched per-layer, per-token gradient method) - Project README with background and experiment plan - Sync script for meru Docker container deployment - Upstream nanoGPT model code copied to src/ Architecture: Computes d(log_p(token))/d(residual_stream) averaged over corpus contexts, replicating Anthropic's Jacobian Lens technique.