datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
tmax-9b-atlas
allenai/tmax-9b Brain Atlas — The Sweet Spot of the Hybrid Family
Cross-post: I ran a brain atlas on the mid-size tmax. Sub-Zero coverage is concentrated in layers 16–30, so read the surgical headroom numbers as a late-layer snapshot.
model: allenai/tmax-9batlas type: activation census + Sub-Zero brain atlas + OV-circuit SVDcorpus: 8,965 promptslayers: 32attention layers: 3, 7, 11, 15, 19, 23, 27, 31hybrid layers: everything elsesacred (fully probed) layers:… See the full description on the dataset page: https://huggingface.co/datasets/juiceb0xc0de/tmax-9b-atlas.qwen3.5-9b-atlas
qwen3.5-9b-atlas
Ornith-1.0-9B-atlas
juiceb0xc0de/Ornith-1.0-9B-atlas
A brain atlas for deepreinforce-ai/Ornith-1.0-9B, the 9B agentic-coding model that reports SOTA results on Terminal-Bench, SWE-Bench, and other agentic coding benchmarks. This is not a chat dataset or a benchmark — it is an internal-mechanics map built by running activations through a corpus of prompts and scoring what each layer, component, head, and feature direction is doing.
If you want to know why this model survives surgical edits, where… See the full description on the dataset page: https://huggingface.co/datasets/juiceb0xc0de/Ornith-1.0-9B-atlas.details_01-ai__Yi-9B-200K
Dataset Card for Evaluation run of 01-ai/Yi-9B-200K
Dataset automatically created during the evaluation run of model 01-ai/Yi-9B-200K.
The dataset is composed of 136 configuration, each one coresponding to one of the evaluated task.
The dataset has been created from 6 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 results.
An additional configuration… See the full description on the dataset page: https://huggingface.co/datasets/OALL/details_01-ai__Yi-9B-200K.qwen-9b-3m
qwen-9b-3m
Multi-domain SFT-target dataset: ~3,000,000 prompts, each with ONE completion
generated offline by Qwen/Qwen3.5-9B (thinking mode). Exported snapshot from an
offline queue pipeline; repartitioned into 512 parquet shards.
Columns (21)
record_index, input_sha256, prompt_sha256, dataset, split, source, upstream_id,
bucket, messages_json, prompt, prompt_token_count, generation_seed, enable_thinking,
worker, executor_worker, completion, completion_input_ids… See the full description on the dataset page: https://huggingface.co/datasets/dipta007/qwen-9b-3m.openwebtext-tokenized-9b
Dataset Card for "openwebtext-tokenized-9b"
More Information needed
Qwen3.5-9B-Base
juiceb0xc0de/Qwen3.5-9B-Base
A brain atlas for Qwen/Qwen3.5-9B-Base, a 32-layer hybrid that runs linear attention on 24 layers and full attention on the other 8. This is not a chat dataset or a benchmark. It is an internal-mechanics map built by running activations through a corpus of prompts and scoring what each layer, component, head, and feature direction is doing.
This is a base model, before any instruction tuning. That makes it a useful thing to have a map of: whatever… See the full description on the dataset page: https://huggingface.co/datasets/juiceb0xc0de/Qwen3.5-9B-Base.c4_codeparrot_3m_3m_mix_9bQwythos-9B-Claude-Mythos-5-1M-atlas
juiceb0xc0de/Qwythos-9B-Claude-Mythos-5-1M-atlas
A brain atlas for empero-ai/Qwythos-9B-Claude-Mythos-5-1M, a 32-layer hybrid that runs linear attention on 24 layers and full attention on the other 8. This is not a chat dataset or a benchmark. It is an internal-mechanics map built by running activations through a corpus of prompts and scoring what each layer, component, head, and feature direction is doing.
The interesting thing about this model is how little of it is full… See the full description on the dataset page: https://huggingface.co/datasets/juiceb0xc0de/Qwythos-9B-Claude-Mythos-5-1M-atlas.harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p05
harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p05
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-30m-kl-0p05,
revision 1bced92ee267198876b452d59e13cf17c41d5df9, job 122412. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
16.90%
mc_logprob
1,780
50.45%
negative_abstain
1,799
41.30%
reversed_qa
1,839
7.94%… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p05.harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p01
harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p01
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-30m-kl-0p01,
revision 8c81d012091313d40b3619f15dec11bf47d19c73, job 122411. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
18.21%
mc_logprob
1,780
50.11%
negative_abstain
1,799
36.85%
reversed_qa
1,839
9.52%… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p01.harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p1
harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p1
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-30m-kl-0p1,
revision d657eca6271c10bdd83a5275f58cf0b3d3dc823d, job 122413. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
15.35%
mc_logprob
1,780
50.22%
negative_abstain
1,799
30.41%
reversed_qa
1,839
6.14%… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-30m-kl-0p1.harvey-closed-book-qwen35-9b-notes-conditioned-5m
harvey-closed-book-qwen35-9b-notes-conditioned-5m
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-5m,
revision 906f2b45e2962698c489cfd1f53ab19a4ebe1458, job 122445. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
7.24%
mc_logprob
1,780
36.85%
negative_abstain
1,799
37.08%
reversed_qa
1,839
2.83%
forward_qa:… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-5m.harvey-closed-book-qwen35-9b-notes-conditioned-10m
harvey-closed-book-qwen35-9b-notes-conditioned-10m
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-10m,
revision 4bf295a036c010040d76960e167bfbf0efd05927, job 122446. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
13.04%
mc_logprob
1,780
41.97%
negative_abstain
1,799
51.70%
reversed_qa
1,839
6.14%
forward_qa:… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-10m.Taur_CoT_Analysis_Project___google__gemma-2-9b-itqwen35-9b-teacher-logits-cacheharvey-closed-book-qwen35-9b-notes-conditioned-100m-kl-0p1
harvey-closed-book-qwen35-9b-notes-conditioned-100m-kl-0p1
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-100m-kl-0p1,
revision 04390b5989d8d9adfe5597339f24a27b9b411c37, job 122416. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
17.06%
mc_logprob
1,780
59.38%
negative_abstain
1,799
27.24%
reversed_qa
1,839
6.42%… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-100m-kl-0p1.harvey-closed-book-qwen35-9b-notes-conditioned-1m
harvey-closed-book-qwen35-9b-notes-conditioned-1m
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-1m,
revision 21c40ef032e9cf968482584191c0b8c269f70001, job 122447. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
5.69%
mc_logprob
1,780
34.61%
negative_abstain
1,799
10.84%
reversed_qa
1,839
2.72%
forward_qa:… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-1m.harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-30m-historical-20t-think
harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-30m-historical-20t-think
1,000 historical evaluation attempts (250 tasks, four samples per task), newly
graded with gpt-5.6-sol using Harvey's original per-criterion rubric prompt
and all-criteria-pass rule. Mean all-pass rate: 5.0000%.
The train split contains evaluation records, not training examples.
Generation and grading protocols
Generation is unchanged: historical 20-turn thinking-enabled
glob/grep/read agent… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-30m-historical-20t-think.details_Slim205__Barka-9b-it_v2
Dataset Card for Evaluation run of Slim205/Barka-9b-it
Dataset automatically created during the evaluation run of model Slim205/Barka-9b-it.
The dataset is composed of 116 configuration, each one coresponding 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 results.
An additional… See the full description on the dataset page: https://huggingface.co/datasets/OALL/details_Slim205__Barka-9b-it_v2.harvey-closed-book-qwen35-9b-notes-conditioned-100m
harvey-closed-book-qwen35-9b-notes-conditioned-100m
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-100m,
revision 0c295885100d6eba4f514752aa081c5b0c73fdec, job 122452. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
31.33%
mc_logprob
1,780
61.85%
negative_abstain
1,799
71.10%
reversed_qa
1,839
19.85%… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-100m.africa-drc-congo-dem-rep-environment-9b29113f
Congo, Dem. Rep. - Environment | Africa (DRC official open data)
4,653 rows - 1 Africa country - 1960-2025 - Repackaged by Electric Sheep Africa
TL;DR
This dataset packages one official CSV resource from DRC as
ML-ready Parquet. The source file is the provenance boundary; all usable
indicators or tabular columns from the resource stay together in this repo.
About the source
Source: Congo, Dem. Rep. - Environment
Publisher: World Bank Group… See the full description on the dataset page: https://huggingface.co/datasets/electricsheepafrica/africa-drc-congo-dem-rep-environment-9b29113f.harvey-closed-book-qwen35-9b-notes-conditioned-30m
harvey-closed-book-qwen35-9b-notes-conditioned-30m
Completed closed-book C&H knowledge evaluation of violetxi/qwen35-9b-harvey-v4-notes-conditioned-30m,
revision add7081dd8969ddbe2f6defbf1997b47031284c1, job 122448. All 7,933 probes completed
without API errors. Each probe category is a separate split in this single repo.
Split
Probes
Accuracy
forward_qa
2,515
22.35%
mc_logprob
1,780
54.38%
negative_abstain
1,799
66.04%
reversed_qa
1,839
12.94%
forward_qa:… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-closed-book-qwen35-9b-notes-conditioned-30m.harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-3m-historical-20t-think
harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-3m-historical-20t-think
1,000 historical evaluation attempts (250 tasks, four samples per task), newly
graded with gpt-5.6-sol using Harvey's original per-criterion rubric prompt
and all-criteria-pass rule. Mean all-pass rate: 1.3000%.
The train split contains evaluation records, not training examples.
Generation and grading protocols
Generation is unchanged: historical 20-turn thinking-enabled
glob/grep/read agent… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-3m-historical-20t-think.harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-10m-historical-20t-think
harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-10m-historical-20t-think
1,000 historical evaluation attempts (250 tasks, four samples per task), newly
graded with gpt-5.6-sol using Harvey's original per-criterion rubric prompt
and all-criteria-pass rule. Mean all-pass rate: 4.0000%.
The train split contains evaluation records, not training examples.
Generation and grading protocols
Generation is unchanged: historical 20-turn thinking-enabled
glob/grep/read agent… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/harvey-eval-gpt56sol-qwen35-9b-notes70-recall30-10m-historical-20t-think.spite-gigaspeech-Euro9B
Spite Dataset
Pseudolabeled speech translation data with quality annotations from multiple metrics. This version uses transcripts from GigaSpeech and translations from EuroLLM-9B-Instruct.
Configs
en_de
en_es
en_fr
en_it
en_ko
en_nl
en_pt
en_ru
en_zh
Usage
from datasets import load_dataset
ds = load_dataset("bpop/spite-CV16-Euro9B", "en_pt")
en_sae_wiki_tokenized_gemma-2-9bopenwebtext_tokenized_gemma-2-9bGenerated with SAELens. OpenWeb text pretokenized with Gemma-2 tokenizer.
{
"sae_lens_version": "3.11.0",
"tokenizer_name": "google/gemma-2-9b",
"original_dataset": "Skylion007/openwebtext",
"original_split": "train",
"original_data_files": null,
"context_size": 1024,
"shuffled": true,
"seed": null,
"begin_batch_token": "bos",
"begin_sequence_token": null,
"sequence_separator_token": "bos"
}
gemma2_9b_it_gsm8k_mcmc_with_promptdetails_Delta-Vector__Odin-9B
Dataset Card for Evaluation run of Delta-Vector/Odin-9B
Dataset automatically created during the evaluation run of model Delta-Vector/Odin-9B.
The dataset is composed of 136 configuration, each one coresponding 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 results.
An additional… See the full description on the dataset page: https://huggingface.co/datasets/OALL/details_Delta-Vector__Odin-9B.