summaryrefslogtreecommitdiff
AgeCommit message (Expand)Author
5 daysAdd tests/test_loss_reweight.py: batch determinism + weighting semantics (run...Void Agent
5 daysloss_reweight: CPU generators for randint/randperm (CUDA generators unsupport...Void Agent
5 daysAdd causal experiments: synthetic frequency-matched pair (Gemini design) + lo...Void Agent
7 daysjlens_v3: chunk VJP probes (16) to fit K2200 4GB; proxy chunk fix; e2e verifi...Void Agent
7 daysAdd faithful J-lens (jlens_v3): W_U-probed residual Jacobian per paper; both-...Void Agent
8 daysGPT-2 J-lens: use local Shakespeare + fix summaryVoid Agent
8 daysGPT-2 Small J-lens: frequency + dimensionality samplingVoid Agent
8 daysAdd Pythia-70m test scriptVoid Agent
8 daysFix: lr→learning_rate in configure_optimizersVoid Agent
8 daysAdd dimensional starvation experiment (d_model=16,32,64,128 vs 384)Void Agent
8 daysAdd controlled frequency ablation experiment (doubled-q)Void Agent
9 daysjlens_v2: fix import path for src/ -> root layoutVoid Agent
9 daysAdd jlens_v2.py: simplified gradient approach for J-lensVoid Agent
9 daysjlens.py: fix hooks — don't detach activations, preserve gradient graphVoid Agent
9 daysjlens.py: fix load_model for dict-format nanoGPT checkpointsVoid Agent
9 daysFix jlens.py import path: src/ -> project rootVoid Agent
9 daysMaxwell GPU fixes: disable bf16 SDPA, force fp32Void Agent
9 daysRestructure: nanoGPT at root, custom code in src/Void Agent
9 daysInitial project setup: J-lens implementation for nanoGPTVoid Agent