Nemotron-Post-Training-v3
Collection
Collection of datasets used in the post-training phase of Nemotron Nano, Super, and Ultra v3. • 51 items • Updated • 148
license stringclasses 1
value | messages listlengths 12 230 | tools listlengths 2 6 | uuid stringlengths 36 36 | used_in listlengths 1 1 | metadata listlengths 0 0 |
|---|---|---|---|---|---|
CC BY 4.0 | [
{
"role": "user",
"content": "Implement a fused matrix addition and scaling kernel.\n\nYour implementation should:\n1. Compute C = (A + B) * scale for square matrices\n2. Use a 2D grid/block layout similar to the seed matrix multiplication kernel\n3. Support any square matrix dimensions\n4. Properly handle ... | [
{
"type": "function",
"function": {
"name": "bash",
"description": "Executes a given bash command in a persistent shell session with optional timeout, ensuring proper handling and security measures.\n\nBefore executing the command, please follow these steps:\n\n1. Directory Verification:\n - I... | d52df9d6-3059-5e01-b54f-5c0ad35bdd85 | [
"ultra_v3"
] | [] |
CC BY 4.0 | [
{
"role": "user",
"content": "Implement a fused set-union-and-scale kernel that:\n1. Computes the union of two sorted integer arrays A and B\n2. Scales each unique element by a given factor\n\nThe algorithm should:\n- Merge elements from arrays A and B in sorted order\n- Keep duplicates (like thrust::set_un... | [
{
"type": "function",
"function": {
"name": "bash",
"description": "Executes a given bash command in a persistent shell session with optional timeout, ensuring proper handling and security measures.\n\nBefore executing the command, please follow these steps:\n\n1. Directory Verification:\n - I... | be0a03f1-84cd-582c-8f5e-b1574213e502 | [
"ultra_v3"
] | [] |
CC BY 4.0 | [
{
"role": "user",
"content": "Implement a CUDA kernel that applies an exponential growth transformation and value clamping to each element.\n\nThe kernel should:\n1. For each element, compute a growth value using: growth = weight * (2.0 * exp(-(input[i] - mu)^2 / sigma^2) - 1.0)\n2. Add the growth to the co... | [
{
"type": "function",
"function": {
"name": "bash",
"description": "Executes a given bash command in a persistent shell session with optional timeout, ensuring proper handling and security measures.\n\nBefore executing the command, please follow these steps:\n\n1. Directory Verification:\n - I... | cf4d16eb-8d1b-5ed0-ae0a-68e107cc65a3 | [
"ultra_v3"
] | [] |
CC BY 4.0 | [
{
"role": "user",
"content": "Implement a row-wise minimum finder kernel that finds both the minimum value\nand its column index for each row in a 2D float array.\n\nYour implementation should:\n1. Use CUB's BlockReduce with ArgMin reducer to find column-wise minima\n2. Launch with one block per row (2D gri... | [
{
"type": "function",
"function": {
"name": "bash",
"description": "Executes a given bash command in a persistent shell session with optional timeout, ensuring proper handling and security measures.\n\nBefore executing the command, please follow these steps:\n\n1. Directory Verification:\n - I... | 30d39f73-f08b-5ceb-a7aa-acf513832948 | [
"ultra_v3"
] | [] |
CC BY 4.0 | [
{
"role": "user",
"content": "Implement a broadcast addition operation that adds two tensors with different shapes.\n\nGiven two tensors:\n- x: shape [x_dim0, 1] (column vector)\n- y: shape [y_dim0, dim1] (2D matrix)\n- output: shape [x_dim0, dim1]\n\nThe operation broadcasts x along dimension 1 and broadca... | [
{
"type": "function",
"function": {
"name": "bash",
"description": "Executes a given bash command in a persistent shell session with optional timeout, ensuring proper handling and security measures.\n\nBefore executing the command, please follow these steps:\n\n1. Directory Verification:\n - I... | d87aabb1-0058-5528-bc46-2453634331e6 | [
"ultra_v3"
] | [] |
CC BY 4.0 | [
{
"role": "user",
"content": "Implement a CUDA kernel that counts the number of pairwise absolute differences\nbetween two arrays that exceed a given threshold.\n\nFor arrays A and B of size n, compute the count of indices i where |A[i] - B[i]| > threshold.\n\nYour implementation should:\n1. Use grid stride... | [
{
"type": "function",
"function": {
"name": "bash",
"description": "Executes a given bash command in a persistent shell session with optional timeout, ensuring proper handling and security measures.\n\nBefore executing the command, please follow these steps:\n\n1. Directory Verification:\n - I... | 45b21d66-363f-57ea-a3c0-2b7003548657 | [
"ultra_v3"
] | [] |
CC BY 4.0 | [
{
"role": "user",
"content": "Compute a matrix of pairwise squared Euclidean distances between two sets of 2D points.\n\nGiven M points in set A and N points in set B, compute an MxN distance matrix\nwhere each element (i,j) contains the squared Euclidean distance:\n\n distances[i * N + j] = (A[i].x - B[j]... | [
{
"type": "function",
"function": {
"name": "bash",
"description": "Executes a given bash command in a persistent shell session with optional timeout, ensuring proper handling and security measures.\n\nBefore executing the command, please follow these steps:\n\n1. Directory Verification:\n - I... | 3ee3d3e0-08a9-55a2-b698-a07e717f552c | [
"ultra_v3"
] | [] |
CC BY 4.0 | [{"role":"user","content":"Implement a vector addition with scaling kernel. The kernel should add tw(...TRUNCATED) | [{"type":"function","function":{"name":"bash","description":"Executes a given bash command in a pers(...TRUNCATED) | 6f48de87-80a2-5cd4-b729-73c05f72dc47 | [
"ultra_v3"
] | [] |
CC BY 4.0 | [{"role":"user","content":"Implement a partitioned prefix sum with chunk index mapping.\n\nYour impl(...TRUNCATED) | [{"type":"function","function":{"name":"bash","description":"Executes a given bash command in a pers(...TRUNCATED) | 395e9e6b-9aa3-54fa-96a5-0f3c47534c1a | [
"ultra_v3"
] | [] |
CC BY 4.0 | [{"role":"user","content":"Implement scaled dot-product attention, a core operation in transformer m(...TRUNCATED) | [{"type":"function","function":{"name":"bash","description":"Executes a given bash command in a pers(...TRUNCATED) | 33d82ca6-6d83-52e8-8111-bc318ebaa03d | [
"ultra_v3"
] | [] |
End of preview. Expand in Data Studio
Dataset Description:
Nemotron-SFT-CUDA-v1 is a training dataset for CUDA code. It helps language models write CUDA kernels and solve CUDA programming problems.
We start from CUDA code in Nemotron Pretraining Code v2, which has a permissive license. An OpenCode agent powered by GLM-4.7 reads that code and writes new CUDA programming problems. Each problem comes with a hidden answer and tests. A second OpenCode + GLM-4.7 agent then tries to solve the problems using only the prompt and a few helper files. We record the full trace and keep only the runs that pass the tests with no Compute-Sanitizer errors.
This dataset is ready for commercial use.
Dataset Owner(s):
NVIDIA Corporation
Dataset Creation Date:
Created on: 2026-03-04
Last Modified on: 2026-03-16
Version:
Nemotron-SFT-CUDA-v1
Previous Version(s): N/A
License/Terms of Use:
This dataset is governed by the Creative Commons Attribution 4.0 International License (CC BY 4.0).
Intended Usage:
This dataset is intended for post-training (supervised fine-tuning) of language models on CUDA tasks. It targets:
- writing CUDA kernels and library calls from a spec;
- reasoning about shapes, memory, indexing, and the CUDA API;
- using compiler errors and test output to fix code.
Dataset Characterization
Data Collection Method
- [Synthetic]
Labeling Method
- [Synthetic]
- [Automated]
Dataset Format
Modality: Text
Format: JSONL
Structure: Text + Metadata
Dataset Quantification
Samples: 2,276
Total Data Storage: ~84 MB.
Reference(s):
Ethical Considerations:
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal developer teams to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse. Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns here
- Downloads last month
- 4