Datasetpublic

cs-giung/nemotron-math-v2-medium-mini

Nemotron Math V2 Medium Mini A compact textual reasoning dataset derived from the medium split of nvidia/Nemotron-Math-v2 at immutable revision 8e793210e175b6406c752a870f585f62de98c0d3. Selection boundary This extraction intentionally excludes tool-use semantics: Accept exactly two messages with roles user then assistant. Reject records declaring tools, containing assistant tool calls, containing tool-result messages, or containing any additional turns. Require… See the full description on the dataset page: https://huggingface.co/datasets/cs-giung/nemotron-math-v2-medium-mini.

sourceHugging Facecc-by-4.0updated 1mo agoView on Hugging Face

0likes13downloads

Dataset Card

Nemotron Math V2 Medium Mini

A compact textual reasoning dataset derived from the medium split of `nvidia/Nemotron-Math-v2` at immutable revision 8e793210e175b6406c752a870f585f62de98c0d3.

Selection boundary

This extraction intentionally excludes tool-use semantics:

—Accept exactly two messages with roles user then assistant.
—Reject records declaring tools, containing assistant tool calls, containing tool-result messages, or containing any additional turns.
—Require the injected source user message to end with the exact problem text.
—Read reasoning only from the sole assistant message's reasoning_content.
—Read the final response only from that assistant message's content.
—Preserve pinned medium-split row order without sampling or shuffling.

Each output row contains id, source, question, steps, and answer. Reasoning is split on blank lines into whitespace-stripped steps. Source UUIDs are not globally unique across trajectories, so every output ID combines the immutable zero-padded source row position with the source UUID.

Rows must contain 3-50 steps, at least 128 total reasoning characters, and no step longer than 1,024 characters. The exact Llama 3 training sequence must be at most 4,096 characters and 1,024 tokens under the prepared models/llama3-1.2b tokenizer.

Dataset statistics

Metric	Value
Source rows scanned	2,502,305
Final records	66,344
Retention	2.651%
File size (canonical JSONL)	172,915,780 bytes
Sources represented	2
Unique questions	25,275
Total reasoning steps	424,750
Mean steps per record	6.40
Median steps per record	6
Steps per record	3-45
Mean individual step length	167.40 characters
Median individual step length	96 characters
Individual step-length range	1-1,024 characters
Median total reasoning length	1,011 characters
Total reasoning-length range	165-3,561 characters
Answer-length range	15-3,006 characters
Question-length range	14-1,222 characters

Source composition

Source	Records
`nvidia/Nemotron-Math-v2/stackflow`	57,892
`nvidia/Nemotron-Math-v2/aops`	8,452
Total	66,344

Rejection accounting

Reason	Rows
Source conversation is not exactly two turns	762,942
Tool trajectory	463,996
Complete Llama 3 sequence over 4,096 characters	404,838
More than 50 reasoning steps	338,489
A reasoning step over 1,024 characters	309,154
Complete Llama 3 sequence over 1,024 tokens	122,303
Fewer than 3 reasoning steps	34,031
User message does not end with exact problem	208
Accepted	66,344
Total	2,502,305

Integrity and provenance

The canonical local JSONL has SHA-256 9840f0263be78e9ee6648287ee1a758f7fd40ee72e6d1012aeae780011fabe4a. The repository includes construction_manifest.json with the pinned source revision, tool-exclusion boundary, ID scheme, complete filter contract, rejection counts, and content hash.

Rows retain the source dataset's row-level Creative Commons licensing. Usage remains subject to the source dataset's licensing terms and documented limitations.