Datasets:
license: other
license_name: mixed-source-licenses
license_link: >-
https://huggingface.co/datasets/KeisukeMiyamoto/SyntheticTalk-jp#source-datasets-and-licenses
license_details: Each row remains subject to the license and terms of its source dataset.
language:
- ja
task_categories:
- text-generation
annotations_creators:
- machine-generated
language_creators:
- found
- machine-generated
multilinguality:
- monolingual
source_datasets:
- APTO-001/instruction-following-dataset
- APTO-001/ja-safety-sft-dataset
- APTO-001/japanese-civil-law-llm-instruction-dataset
- APTO-001/japanese-reasoning-dataset-sample
- APTO-001/japanese-style-contrast-dataset
- APTO-001/llm-safety-japanese-multiturn-dataset
- Aratako/Japanese-Creative-Writing-39.6k
- Aratako/Magpie-Tanuki-8B-97k
- Aratako/Magpie-Tanuki-Instruction-100k-Embeddings
- Aratako/Magpie-Tanuki-Instruction-Selected-Evolved-26.5k
- Aratako/Self-Instruct-Qwen2.5-72B-Instruct-60k
- Aratako/Synthetic-JP-Conversations-Magpie-Nemotron-4-10k
- Aratako/gemma-2-27b-evol-instruct-88k-sft
- CohereLabs/aya_dataset
- DataPilot/Generated-dataset-by-deepseek-v2.5
- DataPilot/Knowledge-QA-SingleTurn-Dataset
- DataPilot/Zero_SFT_Ja_by_Mistral_Small
- DataPilot/Zero_SFT_Ja_v2
- DataPilot/Zero_SFT_Ja_v2_b3t4
- DataPilot/Zero_SFT_Ja_v3.5
- DataPilot/Zero_SFT_Ja_v3_Reasoning
- DeL-TaiseiOzaki/Debatedata_with_feedback
- DeL-TaiseiOzaki/Tengentoppa-sft-qwen2.5-32b-reasoning-100k
- DeL-TaiseiOzaki/Tengentoppa-sft-v4.0
- DeL-TaiseiOzaki/argue_instrcution_data_ja
- DeL-TaiseiOzaki/magpie-gemma-2-9b-it
- DeL-TaiseiOzaki/magpie-qwen2.5-32B-10K-ja
- DeL-TaiseiOzaki/magpie-reasonig-ja-qwen2.5-72b-16k
- OpenAssistant/oasst1
- allenai/WildChat-1M
- llm-jp/extraction-wiki-ja
- llm-jp/llm-jp-chatbot-arena-conversations
- llm-jp/llm-jp-instructions
- llm-jp/magpie-sft-v1.0
- megagonlabs/instruction_ja
- tokyotech-llm/swallow-gemma-magpie-v0.1
models:
- google/gemma-4-31B
- google/gemma-4-12B
pretty_name: LambdaTalk-v2
size_categories:
- 1M<n<10M
tags:
- japanese
- synthetic
- conversational
- multi-turn
- instruction-following
- supervised-fine-tuning
- parquet
configs:
- config_name: default
data_files:
- split: train
path: train-*.parquet
- split: validation
path: validation.parquet
- split: test
path: test.parquet
LambdaTalk-v2
LambdaTalk-v2 is a Japanese synthetic multi-turn conversation dataset generated with Gemma 4 31B. It contains conversations based on seed questions collected from 36 source datasets.
Each conversation contains three user-assistant turns. The first user message is the original seed question. The remaining five messages were generated by Gemma 4 31B.
Purpose
The main purpose of this dataset is supervised fine-tuning of Japanese conversational language models.
Dataset Size
| Split | Rows | User tokens | Assistant tokens |
|---|---|---|---|
| Train | 1,033,187 | 90M | 108M |
| Validation | 10,542 | 919K | 1M |
| Test | 10,542 | 919K | 1M |
| Total | 1,054,271 | 92M | 110M |
Rows were shuffled before the 98% train, 1% validation, and 1% test split. Token counts were measured with the Gemma 4 tokenizer without adding special tokens.
Columns
| Column | Description |
|---|---|
id |
Unique ID. |
source_dataset |
Name of the dataset that contains the seed question. |
messages |
Six-message conversation. Each message has role and content. |
user_tokens |
Total Gemma 4 token count of all user messages in the row. |
assistant_tokens |
Total Gemma 4 token count of all assistant messages in the row. |
Example
from datasets import load_dataset
dataset = load_dataset("<DATASET_REPOSITORY>", split="train")
print(dataset[0]["messages"])
Source Datasets and Licenses
This dataset does not have one uniform license. Each row remains subject to the license and terms of its source dataset. Users must review the linked source terms before modification, redistribution, commercial use, or model training. Unknown means that the permission could not be confirmed from the available license information.
| Source dataset | Rows | License | Modify | Redistribute | Commercial use | Notes |
|---|---|---|---|---|---|---|
| APTO-001/instruction-following-dataset | 155 | apache-2.0 | Yes | Yes | Yes | |
| APTO-001/ja-safety-sft-dataset | 456 | cc-by-4.0 | Yes | Yes | Yes | |
| APTO-001/japanese-civil-law-llm-instruction-dataset | 100 | apache-2.0 | Yes | Yes | Yes | |
| APTO-001/japanese-reasoning-dataset-sample | 65 | apache-2.0 | Yes | Yes | Yes | |
| APTO-001/japanese-style-contrast-dataset | 1,776 | unknown | Yes | Yes | Yes | |
| APTO-001/llm-safety-japanese-multiturn-dataset | 89 | apache-2.0 | Yes | Yes | Yes | |
| Aratako/Japanese-Creative-Writing-39.6k | 15,158 | mit | Yes | Yes | Yes | |
| Aratako/Magpie-Tanuki-8B-97k | 92,365 | apache-2.0 | Yes | Yes | Yes | |
| Aratako/Magpie-Tanuki-Instruction-100k-Embeddings | 92,903 | unknown | Unknown | Unknown | Unknown | |
| Aratako/Magpie-Tanuki-Instruction-Selected-Evolved-26.5k | 25,028 | apache-2.0 + Qwen | Yes | Yes | Yes | |
| Aratako/Self-Instruct-Qwen2.5-72B-Instruct-60k | 58,585 | Apache 2.0 + Qwen License | See terms | See terms | See terms | Subject to the Qwen License. |
| Aratako/Synthetic-JP-Conversations-Magpie-Nemotron-4-10k | 9,224 | apache-2.0 | Yes | Yes | Yes | |
| Aratako/gemma-2-27b-evol-instruct-88k-sft | 80,395 | unknown | Unknown | Unknown | Unknown | |
| CohereLabs/aya_dataset | 5,677 | apache-2.0 | Yes | Yes | Yes | Japanese subset only. |
| DataPilot/Generated-dataset-by-deepseek-v2.5 | 107 | unknown | Unknown | Unknown | Unknown | The referenced DeepSeek model license does not grant rights to data. |
| DataPilot/Knowledge-QA-SingleTurn-Dataset | 5,639 | odc-by | Unknown | Unknown | Unknown | The upstream source license must be checked. |
| DataPilot/Zero_SFT_Ja_by_Mistral_Small | 4,680 | apache-2.0 | Yes | Yes | Yes | |
| DataPilot/Zero_SFT_Ja_v2 | 35,828 | cdla-permissive-2.0 | Yes | Yes | Yes | |
| DataPilot/Zero_SFT_Ja_v2_b3t4 | 13,091 | apache-2.0 | Yes | Yes | Yes | |
| DataPilot/Zero_SFT_Ja_v3.5 | 98,661 | odc-by | Yes | Yes | Yes | |
| DataPilot/Zero_SFT_Ja_v3_Reasoning | 28,490 | odc-by | Yes | Yes | Yes | |
| DeL-TaiseiOzaki/Debatedata_with_feedback | 14 | apache-2.0 | Yes | Yes | Yes | |
| DeL-TaiseiOzaki/Tengentoppa-sft-qwen2.5-32b-reasoning-100k | 103,173 | apache-2.0 | Yes | Yes | Yes | |
| DeL-TaiseiOzaki/Tengentoppa-sft-v4.0 | 159,193 | cc-by-4.0 | Yes | Yes | Yes | |
| DeL-TaiseiOzaki/argue_instrcution_data_ja | 196 | unknown | Unknown | Unknown | Unknown | |
| DeL-TaiseiOzaki/magpie-gemma-2-9b-it | 12,423 | Gemma | Yes | Yes | Yes | |
| DeL-TaiseiOzaki/magpie-qwen2.5-32B-10K-ja | 5,194 | apache-2.0 | Yes | Yes | Yes | |
| DeL-TaiseiOzaki/magpie-reasonig-ja-qwen2.5-72b-16k | 12,034 | unknown | Unknown | Unknown | Unknown | |
| OpenAssistant/oasst1 | 47 | apache-2.0 | Yes | Yes | Yes | Japanese subset only. |
| allenai/WildChat-1M | 1,889 | odc-by | Yes | Yes | Yes | Japanese subset only. |
| llm-jp/extraction-wiki-ja | 29,602 | apache-2.0 | Yes | Yes | Yes | |
| llm-jp/llm-jp-chatbot-arena-conversations | 582 | cc | Yes | Yes | Yes | |
| llm-jp/llm-jp-instructions | 898 | cc-by-4.0 | Yes | Yes | Yes | |
| llm-jp/magpie-sft-v1.0 | 39,003 | apache-2.0 | Yes | Yes | Yes | |
| megagonlabs/instruction_ja | 625 | mit | Yes | Yes | Yes | |
| tokyotech-llm/swallow-gemma-magpie-v0.1 | 120,926 | Gemma | Yes | Yes | Yes |
Notes
This is synthetic data. It may contain factual errors, hallucinations, bias, unsafe content, or low-quality conversations. Users should review the data and source licenses for their own use case.
Support Lambda
Lambda is an open-source project for building small Japanese language models from scratch. As a student, I have funded this project with income from my part-time job, but the growing training costs are becoming difficult to cover.
Your support helps cover GPU costs and develop larger models. Thank you for helping Lambda continue to grow.
Vast.ai
Vast.ai offers affordable cloud GPUs for AI training, with NVIDIA H100 SXM GPUs available from around $1.54 per hour. If you purchase credits through the link below, I receive 3% in GPU credits at no extra cost to you.
https://cloud.vast.ai/?ref_id=521936
Ko-fi
Support Lambda with a donation starting from $5.
