Datasets:

Mesosfer
/

outmind-dataset

text stringlengths 100 1.03M	source stringclasses 1 value	char_len int64 100 1.03M
from autobahn.asyncio.websocket import WebSocketClientProtocol, \ WebSocketClientFactory import asyncio class MyClientProtocol(WebSocketClientProtocol): def onConnect(self, response): print("Server connected: {0}".format(response.peer)) @asyncio.coroutine def on...	code_python	1,264
from itertools import chain from django.utils.itercompat import is_iterable class Tags: """ Built-in tags for internal checks. """ admin = 'admin' caches = 'caches' compatibility = 'compatibility' database = 'database' models = 'models' security = 'security' signals = 'signals'...	code_python	3,104
""" The :mod:`sklearn.utils` module includes various utilites. """ from collections import Sequence import numpy as np from scipy.sparse import issparse import warnings from .murmurhash import murmurhash3_32 from .validation import (as_float_array, check_arrays, safe_asarray, assert_all_fini...	code_python	10,084
""" Python Character Mapping Codec cp1250 generated from 'MAPPINGS/VENDORS/MICSFT/WINDOWS/CP1250.TXT' with gencodec.py. """#" import codecs ### Codec APIs class Codec(codecs.Codec): def encode(self,input,errors='strict'): return codecs.charmap_encode(input,errors,encoding_table) def decode(self,in...	code_python	13,940
# GNU General Public License v3.0+ (see COPYING or https://www.gnu.org/licenses/gpl-3.0.txt) from __future__ import absolute_import, division, print_function __metaclass__ = type ANSIBLE_METADATA = {'metadata_version': '1.1', 'status': ['preview'], 'supported_by': 'community'} ...	code_python	3,887
"import _controls_\nimport new\nnew_instancemethod = new.instancemethod\ndef _swig_setattr_nondynami(...TRUNCATED)	code_python	332,240
"# -- coding: utf-8 --\n\"\"\"\n werkzeug.contrib.cache\n ~~~~~~~~~~~~~~~~~~~~~~\n\n The (...TRUNCATED)	code_python	23,510
"from contextlib import contextmanager\nfrom datetime import datetime\nfrom inspect import isclass\n(...TRUNCATED)	code_python	1,540
"import os\nfrom importlib import import_module\n\nfrom django.core.exceptions import ImproperlyConf(...TRUNCATED)	code_python	8,045
"from __future__ import (absolute_import, division, print_function)\n__metaclass__ = type\n\nimport (...TRUNCATED)	code_python	6,658

End of preview. Expand in Data Studio

YAML Metadata Warning:The task_categories "conversational" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other

OutMind Unified Dataset (Mesosfer-AI)

The official high-quality, multimodal-balanced training dataset curated and cleaned for OutMind, an educational and hackable Small Language Model (SLM) framework developed by Mesosfer-AI.

🌟 Dataset Overview

OutMind dataset is engineered to provide a balanced foundation across Indonesian, English, Chinese, Mathematical Reasoning, and Python Code, organized into three strictly separated tiers:

outmind-dataset/
├── pretrain/              # Cleaned, deduplicated document streams (Parquet)
│   ├── train/             # 98% training split (73 Parquet files)
│   └── val/               # 2% validation split (5 Parquet files)
├── sft/                   # Standardized ChatML multi-turn dialogues (Parquet)
│   ├── train/             # 98% instruction tuning split
│   └── val/               # 2% instruction tuning split
├── validation/            # UNMODIFIED held-out evaluation benchmarks (JSONL)
│   ├── id_tydiqa.jsonl
│   ├── math_gsm8k.jsonl
│   ├── code_humaneval.jsonl
│   └── general_mmlu.jsonl
└── tokenizer/             # Custom BPE Tokenizer engine artifacts
    ├── outmind_tokenizer.json
    └── outmind_tokenizer.model

📚 Data Composition & Sources

1. Pretraining Corpus (~20-25 GB text, partitioned in Snappy Parquet)

Indonesian Wikipedia (id_wikipedia): Complete Indonesian Wikipedia dump, stripped of reference sections, footnote links ([1], [catatan 1]), and wiki table artifacts.
Chinese Wikipedia (zh_wikipedia): Complete Chinese Wikipedia dump normalized to UTF-8 text.
English Cosmopedia v2 (en_cosmopedia): Curated synthetic textbooks and educational stories from HuggingFaceTB/smollm-corpus.
Python Code (code_python): Cleaned Python files from codeparrot/codeparrot-clean, stripped of copyright blocks and license header boilerplate.
Math & Reasoning (math_openwebmath): High-quality mathematical web documents with LaTeX formulas, stripped of web crawler comments and footnote brackets.

2. Supervised Fine-Tuning (SFT / ChatML)

All instruction datasets are unified into the standard ChatML schema:

{
  "messages": [
    {"role": "system", "content": "You are OutMind, an open-source lightweight AI created by Mesosfer-AI."},
    {"role": "user", "content": "..."},
    {"role": "assistant", "content": "..."}
  ],
  "source": "id_aya",
  "num_turns": 3
}

Indonesian: CohereForAI/aya_dataset (Indonesian subset), stripped of legacy <unk> tokens.
Chinese: m-a-p/COIG-CQIA (Chinese open instruction general split).
English: HuggingFaceH4/ultrachat_200k (high-quality multi-turn conversational data).
Math & Reasoning: meta-math/MetaMathQA (GSM8K/MATH augmented solutions).
Code & Logic: ise-uiuc/Magicoder-Evol-Instruct-110K (Python/polyglot code instruction).
Tool Calling: glaiveai/glaive-function-calling-v2 (structured function-calling agent dialogues).
Identity: Handcrafted OutMind & Mesosfer-AI persona grounding.

3. Held-Out Evaluation Benchmarks (Zero Contamination)

Kept completely untouched to evaluate OutMind SLM models fairly:

TyDiQA (Indonesian): Question answering in Indonesian.
GSM8K: Grade school math word problems.
HumanEval: Python programming pass@k benchmark.
MMLU: Broad multidisciplinary 4-choice knowledge evaluation.

🧹 Quality Assurance & Cleaning Standards

This dataset passed the automated OutMind Data-Engineer Audit with an overall quality score of 91.5 / 100 (Grade A - Production Ready):

Footnote & Reference Cleaning: Standalone web citation brackets ([4], [catatan 1]) and Wiki footers (== Referensi ==) cleanly removed.
Copyright Stripping: Top-of-file license comments stripped from code repositories while keeping code bodies intact.
Deduplication: First-turn MD5 hashing applied across SFT dialogues to eliminate identical duplicate prompts.
Array Index Safety: Preserves valid code array indexing like arr[0] without false-positive stripping.

🚀 Quick Start with Hugging Face Datasets

from datasets import load_dataset

# 1. Load Pretraining stream
pretrain_ds = load_dataset("Mesosfer-AI/outmind-dataset", data_dir="pretrain", split="train", streaming=True)
for row in pretrain_ds:
    print(row["source"], len(row["text"]))
    break

# 2. Load SFT Dialogues
sft_ds = load_dataset("Mesosfer-AI/outmind-dataset", data_dir="sft", split="train")
print(f"Total SFT dialogues: {len(sft_ds)}")

⚖️ License

Released under the Apache License, Version 2.0. Individual underlying corpora remain subject to their respective original open licenses.

Downloads last month: 61