datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
omega-het-sft-rl
OMEGA-HET-SFT-RL
Reasoning-trajectory corpora for studying whether heterogeneous (HET) multi-model
SFT data improves post-RL out-of-distribution generalization on OMEGA math vs
homogeneous (HOM) single-model data, under matched controls.
Conditions
HOM: trajectories generated by a single model (Qwen3-4B).
HET: trajectories composed via true token-level continuation across a roster of
7 reasoning models (each model resumes the previous model's own assistant turn… See the full description on the dataset page: https://huggingface.co/datasets/shizhuo2/omega-het-sft-rl.omega-het-expandA-sft
OMEGA-HET-expandA — matched HET-vs-HOM SFT (equal-size)
Matched supervised-fine-tuning data for the OMEGA diversity experiment: for each math prompt,
reasoning trajectories are sampled two ways and only prompts solved (math-verified correct) in both
conditions are kept (matched HOM∩HET = 3,219 prompts), so HET and HOM are directly comparable.
HET (heterogeneous): true token-level continuation across a 3×32B roster
(Qwen3-32B + DeepSeek-R1-Distill-Qwen-32B +… See the full description on the dataset page: https://huggingface.co/datasets/shizhuo2/omega-het-expandA-sft.omega-het-expandA-verified
OMEGA-HET-SFT-RL v2 expandA — VERIFIED (accepted-only) trajectories
Math-verified-correct reasoning trajectories from the expandA pool. HET = heterogeneous 3x32B roster
(Qwen3-32B + DeepSeek-R1-Distill-Qwen-32B + OpenReasoning-Nemotron-32B, true token-level continuation);
HOM = homogeneous single Qwen3-4B. Verified with the OMEGA math verifier (verify.py) against
data/sft_main_prompts.jsonl ground truth; only accepted=true candidates are included here.
Solve rates: HET ~42-45%… See the full description on the dataset page: https://huggingface.co/datasets/shizhuo2/omega-het-expandA-verified.sotopia-omegaomega-het-sft-rl-v2
OMEGA-HET-SFT-RL — v2 dataset (HuggingFace upload)
Repo (target): shizhuo2/omega-het-sft-rl-v2 (public dataset)
Source experiment: does heterogeneous (HET) multi-model SFT reasoning data beat homogeneous (HOM)
single-model data on post-RL OOD generalization (OMEGA math), under matched controls, at diversity
levels {1, K4, ALL}? v2 = English-clean 36-model HET roster, multi-base, OOD-in-RL redistribution.
Contents
1. SFT data (sft_data/) — matched… See the full description on the dataset page: https://huggingface.co/datasets/shizhuo2/omega-het-sft-rl-v2.acornlib-benchmark
Acorn Theorem Proving Benchmark (Preview)
Disclaimer: This is an early-stage, minimal benchmark for internal experimentation. It is not ready for academic publication or production evaluation. The task selection, difficulty calibration, and evaluation methodology are all preliminary.
A small benchmark of 50 theorems from the Acorn proof language, spanning easy to very hard. Each task asks the model to generate a valid proof body that the Acorn verifier accepts.
Data… See the full description on the dataset page: https://huggingface.co/datasets/OmegaCombinator/acornlib-benchmark.bamec66557__VICIOUS_MESH-12B-OMEGA-details
Dataset Card for Evaluation run of bamec66557/VICIOUS_MESH-12B-OMEGA
Dataset automatically created during the evaluation run of model bamec66557/VICIOUS_MESH-12B-OMEGA
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 1 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/bamec66557__VICIOUS_MESH-12B-OMEGA-details.omegacoder-0.3OmegaAgent