datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
made-ternary-suite-8q-20260824-001ternary_quench-traces-datasetTernary-Bonsai-2-27B-Abliterate-LoRA-GGUF-metrics
Ternary-Bonsai-2-27B-Abliterate-LoRA-GGUF-metrics
Measurements behind
AtomicChat/Ternary-Bonsai-2-27B-Abliterate-LoRA-GGUF,
the rank-1 refusal-ablation adapter for PrismML's 1.75 bit/weight ternary pack.
Both packs are covered: every row carries a pack column, PTQ1_0 or PQ2_0. The same
adapter file was run on both, and on the refusal evaluation all 416 greedy replies came out
byte-identical across packs.
Aggregates only. Prompt text is not redistributed (the sources are named in… See the full description on the dataset page: https://huggingface.co/datasets/AtomicChat/Ternary-Bonsai-2-27B-Abliterate-LoRA-GGUF-metrics.ternary_quench-trace-dataset_qwen3_1.7bgrok-1-ternary-quant-experiments
Grok-1 SAAQ quantization / route-preservation experiments
Dataset author: Raul Montoya Cardenas (rmems)
SAAQ stands for Spiking Adaptive Activity Quantization, a term coined by
the dataset author.
Attribution: Grok Build: Grok 4.5 (high) packaged the original 2026-08-10
dataset. Codex: GPT-5.6-Sol (OpenAI) implemented, executed, validated, and published the
canonical issue #85 v4 evidence added on 2026-08-24.
Personal research measuring route preservation when packing open… See the full description on the dataset page: https://huggingface.co/datasets/rmems/grok-1-ternary-quant-experiments.flock
Flock: A negative-enriched protein-protein interaction dataset - Dataset Card
Code: https://github.com/TernaryTx/flock
Overview
Files
file
rows
contents
flock_pairs
404,577
the full dataset: one row per unordered pair, with label, source and leakage flags and examples
leakage_free_pairs
16,210
a benchmark for cofolding models: 335 targets, leakage-filtered
leaked_pairs
14,402
298 targets whose pairs are leaked, as a contrast set… See the full description on the dataset page: https://huggingface.co/datasets/ternarytx/flock.ternary-tinystories-4096
TinyStories 4K-BPE token stream
This is a deterministic, tokenized derivative used by the Ternary LLM Experiment.
The source is
roneneldan/TinyStories.
tokenizer: byte-level BPE, vocabulary 4,096
token dtype: little-endian uint16
training stories: 2,119,719
training tokens: 488,174,163
validation stories: 21,990
validation tokens: 4,907,807
Each story is encoded with explicit beginning/end markers and concatenated into
train.bin or validation.bin. metadata.json records the… See the full description on the dataset page: https://huggingface.co/datasets/sanjuhs/ternary-tinystories-4096.human-annotation-1.5B_judge_preference_ternaryternary-bonsai-memory-footprint-report
Ternary Bonsai — CPU Memory Footprint Report
Ước lượng + kiểm chứng thực tế chi phí RAM (weight + KV-cache) khi chạy các checkpoint
prism-ml/ternary-bonsai
(1.7B / 4B / 8B) trên CPU qua llama.cpp, ở các context length và độ chính xác KV-cache khác nhau.
Dữ liệu thô: memory_data.json. Báo cáo đầy đủ (bảng biểu, giải thích công thức): xem bên dưới.
1. Kiến trúc
Tất cả 3 size dùng head_dim=128, num_key_value_heads=8 (GQA) — nghĩa là 4B và 8B có chi phí
KV-cache/token… See the full description on the dataset page: https://huggingface.co/datasets/tuandunghcmut/ternary-bonsai-memory-footprint-report._judge_preference_ternary_8pairs_judge_preference_ternarytest_api_judge_preference_ternary_judge_preference_ternary_3pairsmpi_ternaryfullchecks_ternary_intransitivity_largehuman-annotation-1.5B_judge_preference_ternary_2_judge_preference_ternary_2pairs_judge_preference_ternary_8pairs_switchfullchecks_ternary_intransitivity