datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
parameter-golf-data
Parameter Golf Competition Data
Pre-tokenized FineWeb shards for the OpenAI Parameter Golf competition. Multiple SentencePiece vocab sizes plus a corrected byte-exact Scylla (TokenMonster) tokenization. Free checkpoint persistence API. Zero setup friction.
⚠️ Important: Scylla v1 Deprecated
The original fineweb_scylla/ directory uses the 998-token vocab from PR #1143. That vocab's byte-accounting metadata treated TokenMonster tokens as context-free, which overcounts… See the full description on the dataset page: https://huggingface.co/datasets/LightSpeedUp/parameter-golf-data.parameter-golf
Parameter Golf FineWeb Export
This repository hosts tokenizer-matched export artifacts derived from HuggingFaceFW/fineweb, specifically a 30B subset pulled from the 100B FineWeb cut used for parameter-golf experiments.
The repository contains:
pretokenized training and validation shards under datasets/datasets/
tokenizer artifacts under datasets/tokenizers/
the export manifest at datasets/manifest.json
selected-document metadata at datasets/docs_selected.jsonl
License… See the full description on the dataset page: https://huggingface.co/datasets/willdepueoai/parameter-golf.Parameter-Golf-V17-512Cube-XYZ-Priority-Orbital-Compression
Parameter Golf V17 — 512Cube Solution Bank
This is an English research-control dataset for OpenAI Parameter Golf work. It is not a replacement for FineWeb and must not be used as a substitute training or validation corpus. FineWeb remains the canonical data path for contest scoring.
The dataset captures three things:
V17 512Cube routing concepts translated into English.
Contest and submission guardrails for legal, reproducible BPB reduction.
Screenshot-derived scouting observations… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V17-512Cube-XYZ-Priority-Orbital-Compression.parameter-golf-sp4096
FineWeb10B SP4096 Dataset for Parameter Golf
This is the SP4096 tokenized version of the FineWeb10B dataset, prepared for the OpenAI Parameter Golf competition.
Contents
fineweb_train_*.bin — Training shards (144 files, ~2.8GB total)
fineweb_val_*.bin — Validation shards (10 files)
tokenizers/fineweb_4096_bpe.model — SentencePiece tokenizer model (4096 vocab)
Format
Binary .bin files contain raw token IDs as uint16 arrays, tokenized with the SP4096… See the full description on the dataset page: https://huggingface.co/datasets/Ryukijano/parameter-golf-sp4096.parameter-golf-sp4096Parameter-Golf-V8-WebSignal-BPE-Entropy-MicroMix
Parameter-Golf-V8-WebSignal-BPE-Entropy-MicroMix
Short Description
Auxiliary micro-mix dataset for OpenAI Parameter Golf: FineWeb-style web-signal filtering, BPE-aware payload compression, boilerplate suppression, privacy-safe minimization, and exact compact-state reasoning for BPB-oriented probes.
Extended Description
This is a V8 auxiliary micro-mix dataset for OpenAI Parameter Golf experiments. It is built from the strongest useful themes in the… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V8-WebSignal-BPE-Entropy-MicroMix.parameter-golf
Parameter Golf FineWeb Export
This is a Parquet-converted fork of willdepueoai/parameter-golf, reformatted for compatibility with the HuggingFace dataset viewer and the datasets library.
The original repository hosts export artifacts derived from HuggingFaceFW/fineweb, specifically a ~10B-token subset pulled from the 100B FineWeb cut used for parameter-golf experiments.
Dataset Structure
train: 15,318,808 documents
validation: 50,000 documents
Each example… See the full description on the dataset page: https://huggingface.co/datasets/mishig/parameter-golf.Parameter-Golf-V15-Patent-Time-Halving
Parameter Golf V15 — Patent Time Halving
English-only working dataset for the 100,000,000-patent runtime audit and 50% time-reduction target.
Status
This package does not claim a measured 0.81 BPB result.
It gives exact calculation rows, audit rows, and Parameter Golf safety rows that can be used for:
clean documentation,
ablation planning,
throughput math,
reproducibility checks,
avoiding invalid val_bpb claims.
Core fixed assumptions
Patent corpus size:… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V15-Patent-Time-Halving.parameter-golf-sp-tokenizers
Parameter Golf SP16384 — Tokenizer + Tokenized FineWeb-10B Shards
SentencePiece BPE tokenizer (vocab_size=16384, byte_fallback=True) + the full FineWeb-10B corpus pre-tokenized with it. Companion artifact to the chaoscontrol submission pipeline; published to make submission-day setup frictionless — no corpus download, no re-tokenization.
Files
Tokenizer (root)
fineweb_16384_bpe.model — SentencePiece model (455 KB).
fineweb_16384_bpe.vocab — Human-readable… See the full description on the dataset page: https://huggingface.co/datasets/Natooka/parameter-golf-sp-tokenizers.Parameter_Golf_V11_Parallel_Residuals_Electronics_Bridge
Parameter Golf V11 — Parallel Residuals Electronics Bridge
This ZIP contains an English training dataset and bilingual explanatory notes for Parameter Golf preparation. It focuses on parallel residuals, serial vs parallel topology, electronics analogies, legal BPB claims, score-first TTT, ablation discipline, RunPod execution, and PR audit behavior.
Important status
This package is educational and preparatory. It does not claim a 0.81 BPB result. It treats 0.81 BPB as an… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter_Golf_V11_Parallel_Residuals_Electronics_Bridge.Parameter-Golf-V7-Resolution-Layout-Compression-Reasoning
Parameter-Golf-V7-Resolution-Layout-Compression-Reasoning
Overview
This is a synthetic auxiliary dataset for OpenAI Parameter Golf style experiments. It is not a replacement for FineWeb. It is designed to test whether a tiny auxiliary mix can reduce wasted next-token probability around compression reasoning, web-page signal filtering, display-vs-payload confusion, and multimodal preview planning.
The central lesson is deliberately narrow:
Smaller rendered objects can… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V7-Resolution-Layout-Compression-Reasoning.parameter-golf-caseops-v1Parameter-Golf-V10-Critical-Memory-FineWeb-MicroMix
Parameter-Golf-V10 Critical Memory FineWeb MicroMix
Version: v10.0.0Language: EnglishFormat: JSONL + critical memory cards + validation scriptsPrimary research objective: build a small, high-signal V10 auxiliary dataset that helps a Parameter Golf workflow move toward an ambitious 0.8 BPB target while preventing false record claims, stale-state drift, and metric mistakes.
Extended description
Parameter-Golf-V10 Critical Memory FineWeb MicroMix is a compact research… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V10-Critical-Memory-FineWeb-MicroMix.parameter_golf_v13_fineweb_systemaware_core_control
Parameter Golf V13 — FineWeb System-Aware Core + Control
This package is an English-first V13 working kit for Parameter Golf. It is designed to be stronger than a pure lesson corpus while still staying honest about what it is and what it is not.
Status
This package does not claim a measured 0.81 BPB result.It treats 0.81 BPB as a research target that still requires:
a legal val_bpb computation,
a reproducible run under the 16 MB artifact cap,
training under the… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/parameter_golf_v13_fineweb_systemaware_core_control.Parameter-Golf-V9-FineWeb-Entropy-Selective-MicroMix
Parameter-Golf-V9-FineWeb-Entropy-Selective-MicroMix
Short description
A tiny auxiliary micro-mix dataset for OpenAI Parameter Golf experiments, focused on FineWeb-like web signal extraction, boilerplate suppression, compact state continuity, BPE-safe plain text, and entropy-selective training probes.
Extended description
Parameter-Golf-V9-FineWeb-Entropy-Selective-MicroMix is designed after the V7/V8 experiments showed that larger synthetic or… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V9-FineWeb-Entropy-Selective-MicroMix.Parameter-Golf-V5
Solutions Training V5
Overview
Solutions Training V5 is a large auxiliary dataset built around one core idea:
the model should not waste probability mass on unnecessary operations, blind guessing, or brute-force symbolic expansion.
V5 was designed mainly from user-provided failure examples.Its central focus is not generic reasoning, but entropy reduction in places where language models often become noisy:
trying to answer without enough context,
forgetting the latest… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V5.Parameter-Golf-V18-V16-V17-Meta-Control-Priority-Bridge
Parameter-Golf V18 — V16/V17 Meta-Control Priority Bridge
This dataset is an extension and bridge for the existing V16 and V17 control-plane datasets.
It combines:
V16 source-ordering and legal discipline corrections.
V17 512Cube XYZ priority routing corrections.
V18 meta-control bridge records connecting both systems into one machine-readable planning layer.
Critical contest note
This dataset is not final contest training data and is not validation data.
Final… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V18-V16-V17-Meta-Control-Priority-Bridge.Parameter-golf-v5x
Solutions Training V5 Extension
Overview
This dataset is a 40,000-example auxiliary extension to Solutions Training V5.
It is built around one central idea:
the model should not brute-force, guess, or over-expand when a symbolic transformation is cleaner and lower-entropy.
This extension was generated primarily from user-provided failure themes:
very large integers ending in 123
cube-volume ×8 scaling rules
shortcut arithmetic instead of repeated expansion
asking for… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-golf-v5x.Parameter_Golf_V12_Thermal_Electronics_BPB_Lesson
Parameter Golf V12 — Thermal Electronics BPB Lesson
Created: 2026-04-22
What this is
V12 is an English dataset and training lesson package for Parameter Golf thinking. It connects:
GPU cooling and thermal throttling;
electronics intuition;
BPB and compression;
RunPod telemetry;
parallel/series/residual mental models;
concise prompt design for structured image generation;
honest record-claim discipline.
What this is not
This is not a new SOTA record claim. It… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter_Golf_V12_Thermal_Electronics_BPB_Lesson.parameter-golf-data
Data Workflows
This directory contains the dataset download helpers and export scripts used for the challenge.
Canonical local layout:
data/datasets/<dataset_name>/
data/tokenizers/
data/manifest.json
data/docs_selected.jsonl
data/docs_selected.source_manifest.json
Downloading Published Data
Download the cached FineWeb export for a tokenizer variant with:
python3 data/cached_challenge_fineweb.py --variant sp1024
This populates ./data/datasets/fineweb10B_sp1024/ and… See the full description on the dataset page: https://huggingface.co/datasets/TCares/parameter-golf-data.Parameter-Golf-V6-Privacy-Web-Filtering
Solutions Training V6 — Privacy Filtering, Unauthorized Access Triage, and Fast Web Signal Extraction
Overview
V6 extends the V5 auxiliary-training idea into a new direction:
the model should learn to jump over noise and sensitive junk instead of reading or repeating everything.
The dataset trains a signal-first behavior for pages, emails, logs, and incident notes:
skip ads, cookie banners, footers, newsletters, and unrelated chrome,
ignore personal-data-heavy… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V6-Privacy-Web-Filtering.Parameter-Golf-V19-Family-Training-Mix
Parameter-Golf V19 — Family Training Mix
V19 is an English training micro-mix package built from the visible 8Planetterraforming dataset family for OpenAI Parameter Golf work.
This package is not just a control note. It contains real training files:
data/train.jsonl
data/validation.jsonl
data/test.jsonl
data/plain_text/train.txt
data/plain_text/v19_micro_0p02pct.txt
data/plain_text/v19_micro_0p05pct.txt
data/plain_text/v19_micro_0p10pct.txt
Core objective
The… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/Parameter-Golf-V19-Family-Training-Mix.parameter_golf_v14_quecto_recovery_tower_english
Parameter-Golf-V14-Quecto-Recovery-Tower-English
English-only auxiliary dataset for compact-model and BPB-oriented experiments.
V14 centers on the compression tower idea: store reusable structure rather than every byte flatly. It also adds the write-delete-recover distinction: training can distill structure into weights, codebooks, caches, residuals, or generators, but exact recovery is possible only when the required information is stored or generated by a reversible rule.
The… See the full description on the dataset page: https://huggingface.co/datasets/8Planetterraforming/parameter_golf_v14_quecto_recovery_tower_english.parameter-golf-fineweb-sp16384
parameter-golf-fineweb-sp16384
SentencePiece BPE tokenizer (vocab_size=16384) and pre-tokenized
FineWeb-10B shards
Structure
canonical/ — used by the submission script. Plain sp16384 BPE
tokenization of FineWeb-10B (50K validation docs + 80 train shards).
caseops/ — not used by the submission. Experiment artifacts from
a CaseOps lossless-case-transform variant.
Submission download
The submission's setup.sh pulls only the canonical/ subdirectory:
hf download… See the full description on the dataset page: https://huggingface.co/datasets/deniskurlov/parameter-golf-fineweb-sp16384.parameter-golfparameter-golf-scyllaparameter-golf
Parameter Golf: Alternative Token Vocabularies
This repository hosts alternative tokenizer vocabularies and matched export artifacts designed for the Parameter Golf challenge. These are experimental variants created to test how different tokenization strategies and vocabulary sizes impact model performance and parameter efficiency.
This work builds directly upon the primary Parameter Golf FineWeb Export by Will DePue (OpenAI). The data herein is derived from the same 30B subset… See the full description on the dataset page: https://huggingface.co/datasets/sysrekt/parameter-golf.parameter-golf-curriculumparameter-golf-sp8192parameter-golf-sp5120