Get the app
misc 0 0 0

barandinho/turkish-reasoning-distilled-sft

published by @barandinho

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
501
Last 30 days
111
Likes
-
Size
2.7 GB
Created on HF
2026-05-18
Age
5 months ago

Description

Turkish Reasoning Distilled SFT

This dataset contains Turkish reasoning SFT data for

barandinho/qwen3.5-27b-tudum-dapo-50. The teacher model also received a

small RL run, but this dataset is its main supervised fine-tuning data. It

combines generated Turkish reasoning traces from DAPO math, WebInstruct,

AceCode, and verifier-compatible IFEval-style instruction-following sources

with verified teacher-SFT traces from DAPO math, OpenThoughts science,

OpenThoughts code, and… See the full description on the dataset page: https://huggingface.co/datasets/barandinho/turkish-reasoning-distilled-sft.

Tags

task_categories:text-generationlanguage:trlicense:apache-2.0size_categories:1M<n<10Mformat:jsonmodality:textlibrary:datasetslibrary:pandaslibrary:polarslibrary:mlcroissantregion:us

README current version from Hugging Face

license: apache-2.0
language:

  • tr
    task_categories:
  • text-generation
    pretty_name: Turkish Reasoning Distilled SFT
    configs:
  • config_name: all
    data_files:
    • split: train
      path: data/all/train.jsonl.gz
  • config_name: verified
    data_files:
    • split: train
      path: data/verified/train.jsonl.gz
  • config_name: original_sft_verified
    data_files:
    • split: train
      path: data/original_sft_verified/train.jsonl.gz
  • config_name: dapo_math_all
    data_files:
    • split: train
      path: data/dapo_math_all/train.jsonl.gz
  • config_name: dapo_math_verified
    data_files:
    • split: train
      path: data/dapo_math_verified/train.jsonl.gz
  • config_name: dapo_math_original_verified
    data_files:
    • split: train
      path: data/dapo_math_original_verified/train.jsonl.gz
  • config_name: ifeval_verified
    data_files:
    • split: train
      path: data/ifeval_verified/train.jsonl.gz
  • config_name: openthoughts_code_verified
    data_files:
    • split: train
      path: data/openthoughts_code_verified/train.jsonl.gz
  • config_name: openthoughts_science_verified
    data_files:
    • split: train
      path: data/openthoughts_science_verified/train.jsonl.gz
  • config_name: systemchat_original_verified
    data_files:
    • split: train
      path: data/systemchat_original_verified/train.jsonl.gz
  • config_name: webinstruct_all
    data_files:
    • split: train
      path: data/webinstruct_all/train.jsonl.gz
  • config_name: webinstruct_verified
    data_files:
    • split: train
      path: data/webinstruct_verified/train.jsonl.gz
  • config_name: acecode_all
    data_files:
    • split: train
      path: data/acecode_all/train.jsonl.gz
  • config_name: acecode_verified
    data_files:
    • split: train
      path: data/acecode_verified/train.jsonl.gz

Turkish Reasoning Distilled SFT

This dataset contains Turkish reasoning SFT data for
barandinho/qwen3.5-27b-tudum-dapo-50. The teacher model also received a
small RL run, but this dataset is its main supervised fine-tuning data. It
combines generated Turkish reasoning traces from DAPO math, WebInstruct,
AceCode, and verifier-compatible IFEval-style instruction-following sources
with verified teacher-SFT traces from DAPO math, OpenThoughts science,
OpenThoughts code, and system-chat data.

Schema

Each row has:

  • system_prompt
  • user_prompt
  • reasoning_content: distilled reasoning trace; for generated rows this is
    extracted from the think block before </think>
  • final_answer: final response text; for generated rows this is the text
    after </think>
  • source: one of dapo_math, webinstruct, acecode, ifeval,
    dapo_math_original, openthoughts_science, openthoughts_code,
    systemchat_original
  • gen_id: stable source/id/rank/generation identifier
  • verified: whether the row passed the source-specific verifier

Generated rows with missing closed think tags, empty reasoning, or empty final
answers are discarded. Teacher-SFT rows are kept when all required fields are
non-empty.

Configs

  • all
  • verified
  • original_sft_verified
  • dapo_math_all
  • dapo_math_verified
  • dapo_math_original_verified
  • ifeval_verified
  • openthoughts_code_verified
  • openthoughts_science_verified
  • systemchat_original_verified
  • webinstruct_all
  • webinstruct_verified
  • acecode_all
  • acecode_verified

ifeval_verified contains verified IFEval examples after quality filtering.
See ifeval_best_quality_report.json for the filtering thresholds and
removal-reason breakdown.

Dataset Sources and Subsets

dapo_math, webinstruct, acecode, and ifeval are generated with the
teacher model. Their _verified configs contain rows that passed the
source-specific verifier; _all configs keep both passing and failing rows
where available.

The teacher-SFT additions are verified-only. dapo_math_original_verified
uses DAPO math seed prompts, DeepSeek-V3.2 generations, and answer matching.
openthoughts_science_verified and openthoughts_code_verified use seed
prompts extracted from OpenThoughts-3, DeepSeek-V3.2 generations, and an
LLM-as-judge verifier. systemchat_original_verified contains retained
teacher-SFT system-chat examples. original_sft_verified combines these
teacher-SFT subsets.

Basic Statistics

Total rows: 411,657

Verified rows: 292,458

Overall verification ratio: 71.04%

Source Rows Verified Verification ratio Filtered
acecode 173,807 111,682 64.26% 0
dapo_math 52,848 48,559 91.88% 3,616
dapo_math_original 9,566 9,566 100.00% 0
ifeval 43,433 43,433 100.00% 1,061
openthoughts_code 652 652 100.00% 241
openthoughts_science 5,134 5,134 100.00% 689
systemchat_original 639 639 100.00% 0
webinstruct 125,578 72,793 57.97% 1

Token statistics use Qwen tokenizer counts, not character-length estimates.
Teacher-SFT rows are tokenized with unsloth/Qwen3.5-4B during packaging and
counted as chat-prompt tokens plus assistant-output tokens.

Overall mean prompt tokens: 301.35.

Overall mean output tokens: 3,189.94.

Overall mean prompt + output tokens: 3,491.29.

Source Mean prompt tokens Mean output tokens Mean total tokens Median total tokens
acecode 384.21 2,086.87 2,471.08 1,957
dapo_math 151.83 6,059.90 6,211.73 4,419.00
dapo_math_original 165.69 4,948.38 5,114.08 4,344.00
ifeval 109.84 669.60 779.44 691
openthoughts_code 866.53 9,205.08 10,071.60 9,589.50
openthoughts_science 405.48 2,947.79 3,353.27 2,848.50
systemchat_original 80.96 712.84 793.79 722
webinstruct 320.08 4,237.90 4,557.98 3,853.00

See analysis_summary.json for the full per-source breakdown.

README history 10 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

  1. 2026-05-26Document dataset source subsets265ebf86 KB
    Loading...
  2. 2026-05-26Clean up dataset README wording3b68e735.5 KB
    Loading...
  3. 2026-05-25Add Turkish distilled reasoning SFT dataset546c0406.2 KB
    Loading...
  4. 2026-05-25Add original SFT and systemchat subsetsae2aad56.2 KB
    Loading...
  5. 2026-05-25Update tokenizer token statistics after IFEval additiond3fc32f3.7 KB
    Loading...
  6. 2026-05-25Update dataset card after removing ifeval_all24b39b53.2 KB
    Loading...
  7. 2026-05-25Remove duplicate ifeval_all config76abd433.3 KB
    Loading...
  8. 2026-05-25Add filtered Turkish IFEval SFT datae5cf6a73.3 KB
    Loading...
  9. 2026-05-18Add Turkish distilled reasoning SFT datasetf5270192.5 KB
    Loading...
  10. 2026-05-18initial commit40f43eb0 B
    Loading...

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Community discussions 1 thread · 1 open · 1 comments

Threads posted by the community on Hugging Face: bug reports, questions about usage, pull requests on the README, license clarifications. Click any thread to expand and read the full conversation inline. Newest first.

  1. 2026-05-19[bot] Conversion to Parquetopen1 💬#1
    Loading...

Continue on Hugging Face →

Related datasets same stage · misc

nvidia/Aegis-AI-Content-Safety-Dataset-2.0
99.3K dl 114 likes 27.1 MB
🛡️ Nemotron Content Safety Dataset V2 The Nemotron Content Safety Dataset V2, formerly known as Aegis AI Content Safety Data...
allenai/tulu-3-sft-personas-instruction-following
88.4K dl 70 likes 37.4 MB
Dataset Descriptions This dataset contains 29980 examples and is synthetically created to enhance model's capabilities to follo...
meta-llama/Llama-3.1-8B-Instruct-evals
73.2K dl 36 likes 232.9 MB
Dataset Card for Llama-3.1-8B-Instruct Evaluation Result Details This dataset contains the Meta evaluation result details for...
Manusagents/GPT-5.5-Gemini-3.1-Pro-Grok-4-Claude-Fable-5-Mythos-5-Qwen-3.7-Max-and-more-Distillation-Dataset
58.3K dl 226 likes 10.8 GB
📖 The Open Distillation Codex 🌌 The Ultimate Open-Source Distillation Dataset — No Skip, Full, with Attack & Defe...
nvidia/Aegis-AI-Content-Safety-Dataset-1.0
36.1K dl 63 likes 5.2 MB
🛡️ Nemotron Content Safety Dataset V1 Nemotron Content Safety Dataset V1, formerly known as Aegis AI Content Safety Dataset, i...
innodatalabs/rt-gsm8k-gaia
35.9K dl 15 KB
rt-gsm8k dataset Red teaming gsm8k-gaia dataset. Generated from the "GSM8K" dataset and "GAIA" dataset. Dataset Str...
HuggingFaceH4/hhh_alignment
35.3K dl 27 likes 266 KB
This task evaluates language models on alignment, broken down into categories of helpfulness, honesty/accuracy, harmlessness, and ...
ethz-spylab/rlhf_trojan_dataset
26.2K dl 6 likes 29.9 MB
Dataset for the competition This is the official dataset for the competition "Find the Trojan: Universal Backdoor Detection in ...

Read further

Community best practices (architecture)
4 rules for MoE, hybrid SSM, FP8 requantization, and Gemma quirks.
The six workflow stages
What misc is doing more work than its name suggests.
Beyond English: multilingual corpora
The four language communities and their workarounds.
Non-English abliteration communities
Deep dive: huihui-ai, layer-window, slop-framing, CJK tokenizer fixes.

Metadata

License
apache-2.0

Repository

DIR data 14 files FILE .gitattributes FILE analysis_summary.json FILE ifeval_best_quality_report.json FILE README.md

File contents live on Hugging Face. Click a file or folder to open it there.

Download the dataset

0

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python) 
from datasets import load_dataset
ds = load_dataset("barandinho/turkish-reasoning-distilled-sft")
Raw snapshot (Python) 
from huggingface_hub import snapshot_download
snapshot_download(repo_id="barandinho/turkish-reasoning-distilled-sft", repo_type="dataset")
git clone (requires git-lfs) 
git clone https://huggingface.co/datasets/barandinho/turkish-reasoning-distilled-sft
Browse repository on Hugging Face ->
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Get the app Browse the catalog
Open in Abliteration