Open-source cleanup: unify on the verified encoder, prune internal docs, add results#11
Merged
Conversation
…cs, add results Bring main onto the verified single-timestep code (the transfer-matrix/ablation code behind the paper's results), while PRESERVING main's extra modules (world_tokenizer/mm_perceiver3.py, viz_suite.py, scripts/) and leaving paper/ untouched. Add committed experiment artifacts: results/ (transfer matrix + ablations), figures/, pixnerd_integration/ (decoder integration), run_ablations.sh. Remove internal-only / redundant docs (HANDOFF, PITCH, PHASE1_EXPLAINED, STATUS_AND_ROADMAP — content lives in PLAN.md / EXPERIMENTS.md). Rewrite top-level README for open source (Kepler-Encoder: what/why, layout, setup, reproduce). Remove dead world_tokenizer/dataloader2.py (superseded by dataloader.py); add a module map to world_tokenizer/README.md.
Yip-Jia-Qi
approved these changes
Jul 13, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Makes
mainthe open-source, contributor-facing codebase.What changed
mainonto the verified single-timestep encoder (the transfer-matrix / ablation code that produced the paper's results). Shared files (train_chunks.py,mm_perceiver2.py,dataloader.py,model.py,dataset.py,visualizer/*) update to that version. main's unique modules are preserved:mm_perceiver3.py,viz_suite.py,scripts/.results/(transfer matrix + ablations),figures/,pixnerd_integration/(decoder integration),run_ablations.sh.HANDOFF,PITCH,PHASE1_EXPLAINED,STATUS_AND_ROADMAP— their content lives inPLAN.md/EXPERIMENTS.md); rewrite the top-levelREADME.mdfor open source; add a module map toworld_tokenizer/README.md.world_tokenizer/dataloader2.py(superseded bydataloader.py).Untouched:
paper/is byte-identical tomain— no conflict with the paper PR (#10); the two can merge in any order.The large diff is expected (code unified to the verified version). Deferred to a follow-up: genericizing hardcoded
/mnt/naspaths + VM-specific notes, and a deeper legacy-code audit.