Nemotron Post-Training Datasets
Collection
Collection of datasets used in the post-training phase of Nemotron Nano, Super, Ultra v3, Super 3.5 VL, and Lightning 3.5. • 64 items • Updated • 209
Datasets:
metadata dict | messages listlengths 2 2 |
|---|---|
{
"uuid": "18a8a6bb-9f40-5fef-8e99-f2097679b113",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 0,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "73d3c3e8-c7ab-598a-8a16-06bb3f88e58a",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 0,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "fb108508-86b3-5abf-8319-ecd2a646064f",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 0,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "925f4f86-50aa-5193-b225-5d0d2df8cb03",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 1,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "777b6bd5-f37a-5ae5-aa73-1b169b8efb85",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 1,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "6a5e49cc-5096-5c37-bcf3-acac090378f3",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 1,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "cf4646ce-8e72-5c1b-8a67-28d501b27355",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 2,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "d179ea62-9dad-5ff5-abf8-163a247ef983",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 2,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "2aeb9298-c9d2-5b32-a5aa-b476f108915f",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 2,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "8a45b840-8831-5c3e-bfec-5fdbb64f5850",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 3,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "7b682439-dff4-52f3-8eaa-c9213f5a776c",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 3,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "1b3b0e81-8355-5810-aee2-5f52d510d1a8",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 3,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "c8fa17ce-af1e-51ae-82e5-998161e26d91",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 4,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "819a773e-7c0f-570e-a0d0-9b775b494ae6",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 4,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "aea102b0-cb67-5be4-987a-2a9971cf7f2f",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 4,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "0f88e0bb-1003-55ad-9310-0bcd8774a2e8",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 5,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "919ce276-748d-5c4e-9715-c09bdb9560f3",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 5,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "adb6e857-09f7-5aeb-bb94-88bcf16bfaa2",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 5,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "6515f4e6-e15c-5f1f-869e-c2cc72b1949b",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 6,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "4c4a8ecb-28b0-5f8a-a98e-57b9ce342da2",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 6,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "23230401-cf3b-5de5-bddb-a95d46b10b58",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 6,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "2caba708-bef8-51cd-85f3-e96da8844b41",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 7,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "377f9d6b-c110-5f5b-b7c1-6a20caecbc16",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 7,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "25bb9331-a75a-5367-9f23-dc59c3be2695",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 7,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "b1e0eb73-ec6c-5d15-9b45-eae7bc03b4bc",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 8,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "57babad6-c6e7-56e5-9b4d-4af8610e74e6",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 8,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "e7b1403e-00e6-5841-80e8-aca93533101a",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 8,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "a44f043e-8efb-51fd-920a-35203928ce14",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 9,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "5fda3f34-d8b3-57ae-bde3-96c02370b1cd",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 9,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "21fb14d3-d891-5513-a06e-91d0047028ad",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 2048,
"generation_ordinal": 9,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
{
"uuid": "73cba4eb-36de-5d71-bddd-44f8a7933e6f",
"dataset": "long-transduction-v10-complete",
"task_type": "unnumbered_streaming_sum",
"target_tokens": 4096,
"generation_ordinal": 0,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nYour task is to output each expression and the result of evaluating the expression.\n\nSee the example below.\n\n```\nInput:\n\n5+6\n2+4-1\n4+8-3+2\n\nOutput:\n\n5+6=11\n2+4-1... |
{
"uuid": "53422d15-27cf-5aa0-bf79-a2d3373d8aea",
"dataset": "long-transduction-v10-complete",
"task_type": "streaming_sum",
"target_tokens": 4096,
"generation_ordinal": 0,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nYour task is to output each expression with its index and the result of evaluating the expr... |
{
"uuid": "f5c0dad7-538b-5eb9-932e-5862e95bfe55",
"dataset": "long-transduction-v10-complete",
"task_type": "shuffled_streaming_sum",
"target_tokens": 4096,
"generation_ordinal": 0,
"seed_namespace": "v10_train"
} | [
{
"role": "user",
"content": "You are a calculator. You will be given a long sequence of simple arithmetic expressions to evaluate.\nEach expression is preceded by a numeric index in brackets like [1], [2], [3], ...\nThe input expressions are SHUFFLED — they appear in arbitrary order, not in numerical order... |
End of preview. Expand in Data Studio
Nemotron-SFT-Long-Context-Transduction-v1
Nemotron-SFT-Long-Context-Transduction-v1 is a synthetic supervised fine-tuning dataset from NVIDIA for training models to track positions and transform sequences over long inputs. It contains 10,000 English instruction-answer pairs across four task families. Inputs and complete reference answers are generated programmatically according to explicit rules; no language model generates the examples or answers.
Tasks
| Task family | Examples |
|---|---|
| Arithmetic evaluation | 2,500 |
| Hexadecimal-token sorting | 2,500 |
| Variable expansion | 2,500 |
| CSV transformations | 2,500 |
| Total | 10,000 |
Task variants include ordered, shuffled, and unnumbered sequences, row and column permutations, and key-to-word lookups. Variable-expansion and CSV lookup tasks use adjective and noun vocabulary from wonderwords.
Prompt generation targets range from 2,048 to 131,072 tokens: 2,048, 4,096, 8,192, 16,384, 32,768, 65,536, and 131,072. These budgets are estimated using cl100k_base; they are generation targets, rather than measured prompt lengths or combined input and output lengths.
Data format
The dataset is provided as a single JSONL file, data/train.jsonl, exposed as the train split. The file contains 10,000 records and occupies approximately 1.22 GB uncompressed.
Each record has two top-level fields:
messages: oneusermessage containing the instruction and synthetic input, followed by oneassistantmessage containing the complete expected answer. Each message hasroleandcontentfields.metadata:uuid,dataset,task_type,target_tokens,generation_ordinal, andseed_namespace. Thedatasetvaluelong-transduction-v10-completeidentifies the generation recipe.
There are no system messages, tool calls, or separate reasoning-content fields. Answers are not packed or truncated.
Loading the dataset
To load the JSONL file after downloading it:
from datasets import load_dataset
dataset = load_dataset("json", data_files="data/train.jsonl", split="train")
Intended use and limitations
The dataset supports supervised fine-tuning and stress-testing for long-context arithmetic, sorting, variable lookup, and CSV transformations. Tasks require complete outputs that follow index, ordering, delimiter, header, and formatting instructions.
Some presentation variants reuse underlying generated material. The dataset is intended for training and is not established as an independent held-out benchmark. Unique record identifiers do not establish content deduplication, and independent validation of every reference answer and benchmark decontamination have not been established.
License
This dataset is licensed under the OpenMDW License Agreement, version 1.1 (OpenMDW-1.1).
Responsible use
Developers should assess whether the dataset meets the requirements of their intended use case. Please report quality, risk, security vulnerabilities, or NVIDIA AI concerns through NVIDIA's reporting channel.
- Downloads last month
- 805