--- pretty_name: Sadik Multilingual Dataset V2 license: cc-by-4.0 task_categories: - text-generation language: - en - hi - ja - zh - de --- # Sadik Multilingual Dataset V2 A 100,000-example synthetic starter corpus for multilingual assistant training. It covers English, Hindi, Hinglish, Japanese, Chinese, and German, with both chat-message and instruction/response formats. ## Files and splits | File | Format | Split | |---|---|---| | `chat_train.jsonl` | Chat messages (`messages`) | Train | | `chat_validation.jsonl` | Chat messages (`messages`) | Validation | | `instruction_train.jsonl` | `instruction`, `input`, `output` | Train | | `instruction_validation.jsonl` | `instruction`, `input`, `output` | Validation | | `dataset_metadata.json` | Dataset metadata | — | The split is approximately 90% train and 10% validation, stratified by language and format. See `dataset_metadata.json` for exact counts. ## Example schemas Chat: ```json {"language":"en","messages":[{"role":"system","content":"You are a helpful assistant."},{"role":"user","content":"Explain machine learning."},{"role":"assistant","content":"Machine learning lets computers learn patterns from examples."}]} ``` Instruction: ```json {"language":"en","instruction":"Explain machine learning.","input":"","output":"Machine learning lets computers learn patterns from examples."} ``` ## Important quality notice These examples are synthetic and programmatically templated. The corpus is useful for testing data loading, formatting, and training pipelines, but it is **not** equivalent to 100,000 independently written or human-reviewed conversations. Many examples share templates and should not be treated as a high-quality general-purpose knowledge corpus. Review, deduplicate, improve linguistic naturalness, and evaluate before using it for serious training or public claims. ## Loading with Hugging Face Datasets ```python from datasets import load_dataset chat = load_dataset("json", data_files={ "train": "chat_train.jsonl", "validation": "chat_validation.jsonl" }) instruction = load_dataset("json", data_files={ "train": "instruction_train.jsonl", "validation": "instruction_validation.jsonl" }) ``` ## Publish on Hugging Face Hub 1. Create or sign in to an account at https://huggingface.co/. 2. Select **New** → **New Dataset**. 3. Choose a repository name such as `sadik-multilingual-dataset-v2`. 4. Choose Public or Private, and select the license only if you have the rights to distribute all content under it. 5. Upload the four JSONL files, this README, and `dataset_metadata.json` to the repository root. 6. Commit the changes and check the dataset preview. 7. Verify the files load with `load_dataset("json", ...)` before announcing the dataset. ### Optional upload from Python ```bash pip install -U huggingface_hub datasets hf auth login ``` ```python from huggingface_hub import HfApi api = HfApi() api.create_repo( repo_id="YOUR_USERNAME/sadik-multilingual-dataset-v2", repo_type="dataset", exist_ok=True ) api.upload_folder( folder_path="PATH_TO_EXTRACTED_DATASET", repo_id="YOUR_USERNAME/sadik-multilingual-dataset-v2", repo_type="dataset" ) ``` Replace `YOUR_USERNAME` and `PATH_TO_EXTRACTED_DATASET` with your own values. Never share your Hugging Face access token. ## License The metadata declares CC-BY-4.0 as a proposed license. Confirm that you own or have permission to distribute the material and that this license is appropriate before publishing.