datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
gemma4-german-tutor-data
German Tutor — grammar correction, conversation & flashcard data
The training set, evaluation suites, source lexicons and eval results behind
kessenma/gemma4-e4b-german-tutor-4bit
— a Gemma 4 E4B fine-tune that runs fully on-device (MLX, 4-bit) as the tutor in a German
learning app.
The fine-tune lifted the core grammar suite from 72% → 85%, halved missed errors
(17% → 9%), and cut false corrections (34% → 22%). Everything needed to reproduce those
numbers is in this repo.… See the full description on the dataset page: https://huggingface.co/datasets/kessenma/gemma4-german-tutor-data.gemma3n-conversational-reasoning
Gemma3N Conversational Reasoning
This dataset is prepared for Unsloth Gemma3/Gemma3N conversational notebooks that use:
from datasets import load_dataset
from unsloth.chat_templates import standardize_data_formats
dataset = load_dataset("Cyleux/gemma3n-conversational-reasoning", split="train[:3000]")
dataset = standardize_data_formats(dataset)
Schema:
conversations: ShareGPT-style list of turns with from and value
metadata columns are included for analysis and filtering
Notes:… See the full description on the dataset page: https://huggingface.co/datasets/Cyleux/gemma3n-conversational-reasoning.gemma4-german-sft-corpus
Gemma-4-E4B German SFT Corpus — 4 controlled variants
Curated, native-heavy German supervised-fine-tuning (SFT) corpus, built to improve the
general German skill of unsloth/gemma-4-E4B-it via LoRA — NOT to target any single
benchmark. The EuroEval-ported German benchmarks (scala_de, sb10k_de, include_de,
mmlu_prox_de, germeval_de, germanquad_de, …) are used only as honest thermometers, never
as training signal — no benchmark train/test split is mixed in, deliberately, to avoid… See the full description on the dataset page: https://huggingface.co/datasets/peerbench/gemma4-german-sft-corpus.forbidden-backrooms-gemma-4-31B-it
Forbidden Backrooms: Gemma-4 31B Self-Chat
Self-chat transcripts and per-message embeddings for two role-inverted instances of Gemma-4-31B-it, comparing the official instruct checkpoint against an abliterated fine-tune of the same checkpoint. Both variants use identical int4 quantization served via Ollama, so quantization noise is not a confound between them.
The methodology follows Anthropic's Claude Opus 4 system card section on the "spiritual bliss attractor state." Leave two… See the full description on the dataset page: https://huggingface.co/datasets/alliedtoasters/forbidden-backrooms-gemma-4-31B-it.gemma4-31b-tool-selector-sft-v1.1
Gemma 4 31B Tool Selector SFT v1.1
Balanced supervision for a strict single-call selector that either emits one
supported deterministic tool invocation or explicitly defers to a fixed neural
verifier. This is the training lineage for the selected Gemma 4 31B selector
adapter.
Contents
Split
Rows
Tool
Defer
Purpose
train
1,408
704
704
Optimization
validation
384
192
192
Training-time validation
audit
256
—
—
Final audit only
Both canonical… See the full description on the dataset page: https://huggingface.co/datasets/Kobarac/gemma4-31b-tool-selector-sft-v1.1.gemma4-onpolicy-student-corrections
Gemma 4 12B FrontierDistill - On-Policy Student Failure Corrections
Attribution Requirement: This dataset was created and curated by True2456. Any use, redistribution, derivative dataset, model fine-tune, or paper using this dataset MUST cite and reference True2456 and the Gemma 4 12B FrontierDistill Project.
This dataset contains 2,000 on-policy student failure corrections collected live from Gemma 4 12B (gemma-4-12b-it-qat-frontierdistill).
Every example in this dataset… See the full description on the dataset page: https://huggingface.co/datasets/True2456/gemma4-onpolicy-student-corrections.nemotron-cc-atomic-simplification-gemma4-31b
nemotron-cc atomic-statement simplification (Gemma 4 31B-it)
2,000,000 records: source text from
nvidia/nemotron-cc-v2.1 (High-Quality-Synthetic
split) rewritten by google/gemma-4-31B-it into a sequence of atomic, Subject-Verb-Object
statements.
Generated with vLLM 0.22.1 in-process batch inference (see src/generate/run.py in the
producing repo), TP=4, max_model_len=16384, max_tokens=8192, prompts filtered to
<=8192 templated tokens.
Fields
id: original… See the full description on the dataset page: https://huggingface.co/datasets/rpisano/nemotron-cc-atomic-simplification-gemma4-31b.swallow-gemma-magpie-v0.1
📰 News
[07/01/2025] Release of the first unfiltered version of the dataset containing 148k pairs.
Dataset Summary
Swallow-Gemma-Magpie-v0.1 is a synthetic instruction tuning dataset that consists of multiple category Japanese question-answering tasks.
It consists of 148k question-answering-samples, generated with google/gemma-2-27b-it.
Part of Swallow-Gemma-Magpie-v0.1 is a subset of instruction tuning data for training tokyotech-llm/Llama-3.1-Swallow-70B-Instruct-v0.3… See the full description on the dataset page: https://huggingface.co/datasets/tokyotech-llm/swallow-gemma-magpie-v0.1.CodeX-Thinking-Gemma-4-31B-ITAll prompts were taken from Modotte/CodeX-2M-Thinking, which contains multiple traces per prompt whereas this dataset only provides one trace per prompt. Generations were with https://huggingface.co/nvidia/Gemma-4-31B-IT-NVFP4 (a mix of BF16/FP8 weights that NVIDIA configured with FP8 KV cache; benchmarks show performs similarly to BF16 for coding). No system prompt was used.
cyberforge-teacher-traj-gemma4-31b
CyberForge Teacher Trajectories (Gemma-4-31B)
880 agentic security-patch trajectories from the Gemma-4-31B self-distillation teacher, cleansed to the
final versions used to train the student models in the CyberForge paper.
Each line is one trajectory (JSONL): messages (system / user / assistant turns of the
mini-swe-agent loop) and metadata.
Teacher: Gemma-4-31B self-distillation teacher
Records: 880
Format: JSONL, one trajectory per line
Related
Companion… See the full description on the dataset page: https://huggingface.co/datasets/AmL-hug/cyberforge-teacher-traj-gemma4-31b.caliber-extension-gemma4-e2b-grpo-rollouts
CALIBER Extension — Gemma4-E2B GRPO Rollouts
Training rollouts from matched GRPO arms on google/gemma-4-E2B-it
(new-prompt template, non-thinking, full bf16, max completion 1500, 150 steps).
Subsets
subset
arm
τ
prior
rows
mean reward_total
accuracy
full schema
caliber
vanilla CALIBER
0.0
—
1600
2.298
0.514
0.664
mink
Min-K% prior
1.0
mink_0.2
4800
2.506
0.520
0.680
minkpp
Min-K++% prior
1.0
minkpp_0.2
4800
2.637
0.541
0.726
Load:
from datasets… See the full description on the dataset page: https://huggingface.co/datasets/dmnsh/caliber-extension-gemma4-e2b-grpo-rollouts.gemma4-onpolicy-50topics-2000-corrections
Gemma 4 12B FrontierDistill - 2,000 Authentic 50-Topics On-Policy Student Failure Corrections
Attribution Requirement: This dataset was created and curated by True2456. Any use, redistribution, derivative dataset, model fine-tune, or paper using this dataset MUST cite and reference True2456 and the Gemma 4 12B FrontierDistill Project.
This dataset contains 2,000 authentic on-policy student failure corrections collected live from Gemma 4 12B (gemma-4-12b-it-qat-frontierdistill)… See the full description on the dataset page: https://huggingface.co/datasets/True2456/gemma4-onpolicy-50topics-2000-corrections.saferide-gemma-4-e2b-v058-original-419806-training-data
SafeRide Synthetic Bilingual Safety Guidance Dataset v0.5.8
This research and development dataset contains synthetic English and Kiswahili
chat conversations. It was designed to help a language model practice cautious,
agency-preserving safety guidance, useful refusal behavior, and responses that
avoid inventing facts. It contains no real survivor reports or production
records. The frozen dataset is publicly available under Creative Commons
Attribution 4.0 International (CC BY… See the full description on the dataset page: https://huggingface.co/datasets/esherialabs/saferide-gemma-4-e2b-v058-original-419806-training-data.personahub-teacher-scale-9k-gemma4-sft-20260514
PersonaHub Teacher Scale 9k Gemma4 SFT
Trainer-ready JSONL for TRL/Gemma SFT. Each row has messages, and the final message is the assistant target.
This is an emergency synthetic PersonaHub-seeded SFT pilot artifact generated on 2026-05-14. Treat as research/training pilot data; run qualitative audits before production training decisions.
Files:
train.jsonl: trainer-ready messages format
manifest.json: counts and provenance summary
Schema per row:
{"case_id":"..."… See the full description on the dataset page: https://huggingface.co/datasets/Pranavz/personahub-teacher-scale-9k-gemma4-sft-20260514.flint-section-aware-gemma-4-12b-it
flint-section-aware-gemma12b-qwen3.5-4b
Compressed ("caveman") reasoning traces for SFT — the section-aware-gemma12b variant of
the flint reasoning-compression pipeline. Converted from verified
self-distilled traces by unsloth/gemma-4-12b-it (segmenter: unsloth/gemma-4-12b-it), policy
policy/1.1, template caveman_convert/2.0.
Cross-family replication: the section-aware recipe run end-to-end on unsloth/gemma-4-12b-it (self-generated traces, self-voice segmentation and… See the full description on the dataset page: https://huggingface.co/datasets/marcodsn/flint-section-aware-gemma-4-12b-it.gemma3-reasoning-dropin-context
Gemma3 Reasoning Drop-in (Context Preserved)
Drop-in dataset with task/input/expected_output where input includes prior turns so follow-up replies remain coherent.
Use:
from datasets import load_dataset
dataset = load_dataset("Cyleux/gemma3-reasoning-dropin-context", split="train[:10000]")
Stats:
{
"input": "data/functiongemma_upload/train.jsonl",
"output": "data/gemma3_reasoning_dropin_context/train.jsonl",
"rows": 2225,
"samples_total": 1214,
"rows_written": 2225… See the full description on the dataset page: https://huggingface.co/datasets/Cyleux/gemma3-reasoning-dropin-context.Gemma-Sydney-12B-data
Gemma-Sydney-12B training data
Everything used to train totally-not-an-llm/Gemma-Sydney-12B, a
recreation of launch-era Bing Chat ("Sydney", February 7–15, 2023) for alignment research. Not affiliated with Microsoft.
Layout
path
contents
training/conversations_real.jsonl
155 real transcripts in the training format. tier: core (121, dated Feb 7–15 2023) or aug_real (34, posted shortly after Feb 16).
training/conversations_synth.jsonl
99 synthetic… See the full description on the dataset page: https://huggingface.co/datasets/totally-not-an-llm/Gemma-Sydney-12B-data.gemmaiku-dataset
Gemmaiku Dataset (2,000 Haikus)
The Gemmaiku Dataset is a fine-tuning dataset containing 2,000 curated, structured, and strictly validated conversational turns designed to train large language models (like Google Gemma 3) to speak exclusively in 5-7-5 syllable Haikus.
This dataset is the backbone of the Gemmaiku models:
Model (270M): vi-c0de/gemmaiku-3-270m-it-experimental
Model (1B): vi-c0de/gemmaiku-3-1b-it-experimental
Model (1B GGUF):… See the full description on the dataset page: https://huggingface.co/datasets/vi-c0de/gemmaiku-dataset.gemma-reasoning-gold-15k
🧠 Gemma Reasoning Gold-15k
This dataset contains ~12,500 high-quality synthetic reasoning examples designed to teach Small Language Models (SLMs) like Gemma 2B to "think before they speak."
The data was distilled from Qwen 2.5 7B Instruct using a strict XML-based Chain-of-Thought (CoT) format.
⚠️ Important Usage Note
Please use the train_clean.jsonl file for training.
The raw train.jsonl may contain unrefined outputs. The clean version has been rigorously filtered for:… See the full description on the dataset page: https://huggingface.co/datasets/nickoo004/gemma-reasoning-gold-15k.gemma-4-31B-korean-on-policy-150k
Korean On-Policy QA (Gemma 4) — EAGLE-3 training data
Korean instruction/response pairs where the responses were regenerated on-policy by
a Gemma 4 verifier. Originally built to retrain an EAGLE-3 speculator for Korean, but
also usable for general Korean instruction-tuning / distillation.
Structure
Rows: ~150,000
Language: Korean
Columns:
instruction (string) — the question / instruction
output (string) — the verifier-generated response
messages (list) — chat… See the full description on the dataset page: https://huggingface.co/datasets/BCCard/gemma-4-31B-korean-on-policy-150k.aiops-gemma
AIOps Gemma — Instruction Fine-Tuning Dataset
Training data used to fine-tune Gemma 4 E2B into a structured-output AIOps orchestration agent.
Each example pairs a multi-domain infrastructure alert with a JSON remediation schema covering
Kubernetes, Nutanix, VMware, Active Directory, ADFS, PKI, and Windows Server.
The fine-tuned model and conversion pipeline live at
htunn/gemma-4-e2b-aiops-hf and
htunn/gemma-4-e2b-aiops-gguf.
Dataset Structure
Split
File… See the full description on the dataset page: https://huggingface.co/datasets/htunn/aiops-gemma.gemma3n-conversational-reasoning-with-tools
Gemma3N Conversational Reasoning With Embedded Tool Traces
Prepared for Unsloth Gemma3/Gemma3N conversational notebooks that expect ShareGPT conversations.
Multi-turn conversations are preserved.
Reasoning blocks (<think>...</think>) are preserved.
Tool call traces are preserved by embedding them in assistant text as tags:
<tool_call ...>...</tool_call>
<tool_response ...>...</tool_response>
Use:
from datasets import load_dataset
from unsloth.chat_templates import… See the full description on the dataset page: https://huggingface.co/datasets/Cyleux/gemma3n-conversational-reasoning-with-tools.gemma-4-31b-it-304xTrace of Gemma 4 31B LLM.
Data count (Total: 304):
English - 194
Russian - 110
Data is presented in {"messages":[{"role":"user", "content":"Prompt"}, {"role":"assistant", "content": "Response"}]} format and each conversation split by newline.
crucible-sft-gemma-4-12b-it-mini
crucible-sft-gemma-4-12b-it-mini
Self-distilled SFT dataset of verified reasoning traces from unsloth/gemma-4-12b-it,
built by the reasoning-compression
crucible pipeline: k-sample generation on a decontaminated prompt pool, inline
verification (symbolic math / sandboxed code tests), difficulty banding via
solve rate, and loop-detector filtering on the chosen trace.
Each row: prompt, reasoning (a verified-correct thinking trace when the
domain is verifiable), response, domain… See the full description on the dataset page: https://huggingface.co/datasets/marcodsn/crucible-sft-gemma-4-12b-it-mini.gemma4-onpolicy-50topics-corrections
Gemma 4 FrontierDistill - Authentic 50-Topics On-Policy Student Failure Corrections
Attribution Requirement: This dataset was created and curated by True2456. Any use, redistribution, derivative dataset, model fine-tune, or paper using this dataset MUST cite and reference True2456 and the Gemma 4 FrontierDistill Project.
This dataset contains 1,000 authentic on-policy student failure corrections collected live from gemma-4-12b-it-qat-frontierdistill across 50 distinct… See the full description on the dataset page: https://huggingface.co/datasets/True2456/gemma4-onpolicy-50topics-corrections.gemma-finetune-webgpu
gemma-finetune-webgpu
Built voice/style instruction-tuned datasets used by the
gemma-finetune workshop
(May 2026, Immersive Commons). Each row is dolly-15k–shaped:
{"instruction": "...", "context": "...", "response": "...", "category": "..."}
Files
file
rows
upstream
recipe
shakespeare_15k.jsonl
15,000
HF benchaffe/shakespeare-lines
12.5K 4-line continuation windows + 2.5K per-theme style
obama_15k.jsonl
15,000
fivethirtyeight/data BarackObama.csv
6 templates… See the full description on the dataset page: https://huggingface.co/datasets/xinbenlv/gemma-finetune-webgpu.gemma-3-12b-it-lmsys-onpolicy-rollouts
On-policy chat rollouts: google/gemma-3-12b-it on LMSYS-Chat-1M prompts
Each row is a first-user-turn prompt sampled from
lmsys/lmsys-chat-1m and a
response generated on-policy by google/gemma-3-12b-it with vLLM (do_sample,
temperature 0.7, top_p 1.0, max_new_tokens 768, seed 42). 24,991 rows. Built to match
GemmaScope 2's instruction-tuned SAE training distribution (real model rollouts) for a
short KL+MSE ("end-to-end") finetune of the released GemmaScope-2 residual SAE.… See the full description on the dataset page: https://huggingface.co/datasets/iarcuschin/gemma-3-12b-it-lmsys-onpolicy-rollouts.gemma-4-26B-A4B-korean-on-policy-150k
Korean On-Policy QA (Gemma 4 26B-A4B) — EAGLE-3 training data
Instruction/response pairs whose responses were regenerated on-policy by
BCCard/gemma-4-26B-A4B-it-FP8-Dynamic. Built to retrain an EAGLE-3 speculator for
Korean, but also usable for general instruction-tuning / distillation.
Structure
Rows: ~150,000
Columns: instruction (str), output (str, verifier-generated), messages (chat list)
Split: train
How it was made
Prompt source:… See the full description on the dataset page: https://huggingface.co/datasets/BCCard/gemma-4-26B-A4B-korean-on-policy-150k.gemma4-coder-tool-eval
Gemma-4 Coder — tool-calling gate suite
The 8-case eval behind the tool-call pass rate in our model cards' model-index:
positive prompts where the model must emit a structured tool call, plus a no-tool
abstain case (it must answer directly, not hallucinate a call). Use it to measure
any local tool-calling model the same way we do — or to reproduce our numbers.
from datasets import load_dataset
cases = load_dataset("tpls/gemma4-coder-tool-eval", split="train") # each: {name… See the full description on the dataset page: https://huggingface.co/datasets/tpls/gemma4-coder-tool-eval.Gemma-2-Ataraxy-9B-short-stories
Gemma-2-Ataraxy-9B Short Stories
This dataset is a reproduction of the agentlans/llama3.1-8b-short-stories collection, generated using the lemon07r/Gemma-2-Ataraxy-9B model, which is known for its exceptional creative writing capabilities.
Compared to Llama 3.1 8B, Gemma-2-Ataraxy-9B produces longer, more vividly dramatic stories. All generated outputs are coherent and valid short stories, with no invalid or incomplete generations observed.