Datasets:

N8Programs
/

shared-emergence-icl-modalities-128

pretty_name: Shared emergence ICL modalities - 128 trials per task
task_categories:
  - text-generation
tags:
  - in-context-learning
  - reproducibility
  - multimodal

Shared-emergence ICL replication at T=128

This dataset contains the complete raw result archive for the paper “Many Next-Token Predictors are In-Context Learners.”

The campaign evaluates a fixed suite of 100 program-synthesis tasks using 128 sampled prompts per task, for every clean and deranged shot cell described by the paper:

21 run keys;
281 experiment cells;
12,800 predictions per cell;
3,596,800 predictions in total.

The archive expands to a top-level results_128/ directory. It contains raw per-task and per-trial JSON records, including predictions, exact-match correctness, edit distances, and run configuration metadata. Summary JSON files are present for the Evo2 runner but are not counted as experiment cells.

Use the code, complete run manifest, exact model revisions, strict validator, analysis scripts, and per-file hashes in N8python/shared-emergence-icl-modalities.

Integrity

Archive: icl_128_results_2026-07-28.tar.gz

SHA-256: aa4c7b5331092582b14ce7cc1c025ef55f4a05a26696299cd69973e8da942931

Compressed size: 718,954,210 bytes.

The repository fetcher verifies both the byte size and SHA-256 before extracting, and the validator checks the complete 281-cell grid and all 3,596,800 trial records.