datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
llama-cpp-wheelsIf you like this please consider liking and donating (https://buymeacoffee.com/aiencoder)
🏭 llama-cpp-python Mega-Factory Wheels
"Stop waiting for pip to compile. Just install and run."
The most complete collection of pre-built llama-cpp-python wheels in existence — 8,333 wheels across every platform, Python version, backend, and CPU optimization level.
No more cmake, gcc, or compilation hell. No more waiting 10 minutes for a build that might fail. Just find your wheel and… See the full description on the dataset page: https://huggingface.co/datasets/AIencoder/llama-cpp-wheels.qwen-cpp-agent-0-protocolExperiment in agentic autonomy protocols.
~ everything in this repo was created by Qwen 3.8 27B (Q4) running autonomously inside Deepseek Harness, on a single RTX 3090 GPU, for 3 weeks.
The only human artifacts are:
agents/*
human/*
AGENTS.md
Cpp-Code-LargeCpp-Code-Large
Cpp-Code-Large is a large-scale corpus of C++ source code comprising more than 5 million lines of C++ code. The dataset is designed to support research in large language model (LLM) pretraining, code intelligence, software engineering automation, and static program analysis for the C++ ecosystem.
By providing a high-volume, language-specific corpus, Cpp-Code-Large enables systematic experimentation in C++-focused model training, domain adaptation, and downstream code… See the full description on the dataset page: https://huggingface.co/datasets/ajibawa-2023/Cpp-Code-Large.code_contest_instruct_cppcpp-code-code_search_net-style
C++ Dataset
documentation source: https://huggingface.co/docs/datasets/main/en/repository_structure
Supported Tasks and Leaderboards
language-modeling: The dataset can be used to train a model for modelling programming languages, which consists in building language models for programming languages.
Language
C++ programming language
Dataset Structure
Data Instances
A data point consists of a function code along with its documentation.… See the full description on the dataset page: https://huggingface.co/datasets/malteklaes/cpp-code-code_search_net-style.cpp_cwe_GRPO
cpp_cwe_GRPO
VeRL/GRPO-ready C++ security-coding dataset produced from every harness-passing stage-6 rewrite in the simple_gen pipeline.
Files
File
Rows
Description
cpp_cwe_GRPO.parquet
8,039
Full corrected dataset
cpp_cwe_GRPO_train.parquet
7,236
Deterministic 90% training split
cpp_cwe_GRPO_val.parquet
803
Deterministic 10% validation split
The split uses seed 20260929. Each record has a unique function name, and the two splits are disjoint.… See the full description on the dataset page: https://huggingface.co/datasets/AetherPrior/cpp_cwe_GRPO.leetcode-codegen-cpp
LeetCode Code-Gen Dataset — C++
4025 rows. Given a problem statement, its input/output examples, and a
required algorithm/technique, generate a correct C++ solution.
Part of a 4-language collection built from the same source: see the sibling
Python,
Java,
C++, and
JavaScript
datasets.
Verification
Not execution-verified. There is currently no compiler/runtime harness for this language in the build pipeline (only Python has one). Rows are extracted directly from… See the full description on the dataset page: https://huggingface.co/datasets/AmareshHebbar/leetcode-codegen-cpp.cpp-project
中文 C++ 项目编程训练集
由 100 份中文 C++ 控制台项目语料转换而来，涵盖管理系统、实用工具、小游戏和基础算法。每个项目包含题目、原有解题思路和完整代码。适用于中文代码生成和指令微调.Author : KochiyaSanaeNya
数据格式
三个配置是100 条样本的不同表示,训练时选择其中一种.所有文件均为 UTF-8 无 BOM JSONL .
配置
训练字段
训练集
验证集
sft（默认）
messages：user / assistant 对话
90
10
alpaca
instruction / input / output
90
10
pretrain
text：包含三个原始章节的全文
90
10
每条记录还包含 id、title、source_file 和 source_sha256，用于追溯来源，不需要作为训练输入。
SFT 用户消息包含 C++17 作答要求和原题；助手消息包含原有解题思路及 Markdown C++ 代码块。
Alpaca 的… See the full description on the dataset page: https://huggingface.co/datasets/AsyncLab/cpp-project.smollm3-stack-v2-Cpp
Synthetic Pre-pretraining Datasets
This dataset contains the pre-processed synthetic pre-pretraining (PPT) and pre-training (PT) data used in the paper Synthetic Pre-pretraining Survives Scale, but Not as a Grammatical Prior.
It includes a range of PPT tasks (e.g., k-Shuffle Dyck, Set, MP-Struct Core, NCA) and standard pre-training mixtures (e.g., C4, SmolLM3, Olmo3, Marin) used to evaluate PPT at scale.
Code: GitHub repository
Project page: Hugging Face Organization
For a… See the full description on the dataset page: https://huggingface.co/datasets/verify-ppt/smollm3-stack-v2-Cpp.cpp-10k10k random lines of the "text" column of the https://huggingface.co/datasets/wttw/code_contest_instruct_cpp dataset
glm47-pie-cpp-posttraining-data
GLM-4.7-Flash PIE C++ Post-Training Data
The exact prepared dataset used for the GLM-4.7-Flash C++ performance
post-training runs.
Splits
File
Rows
Purpose
sft/train.jsonl
7,864
Supervised fine-tuning
grpo/train.jsonl
7,887
GRPO prompt and reward evaluation
eval/validation.jsonl
1,259
Full held-out evaluation
eval/validation_mini126.jsonl
126
Fast evaluation
eval/validation_mini4.jsonl
4
Smoke evaluation
tasks.tar.gz
9,146 task JSONs
Reward… See the full description on the dataset page: https://huggingface.co/datasets/TokenBender/glm47-pie-cpp-posttraining-data.vgen_cpp
Dataset Card for Opencores
In the process of continual pre-training, we utilized the publicly available VGen dataset.
VGen aggregates Verilog repositories from GitHub, systematically filters out duplicates and excessively large files, and retains only those files containing \texttt{module} and \texttt{endmodule} statements.
We also incorporated the CodeSearchNet dataset \cite{codesearchnet}, which contains approximately 40MB function codes and their documentation.… See the full description on the dataset page: https://huggingface.co/datasets/LLM-EDA/vgen_cpp.CPP-Code-Solutions
C++ Code Solutions
Features
1000k of Python Code Solutions for Text Generation and Question Answering
C++ Coding Problems labelled by topic and difficulty
Recommendations
Train your Model on Logical Operations and Mathematical Problems Before Training it on this. This is optional for Fine Tuning 2B parameter + models.
Format the prompts in a orderly way when formatting data eg. {question} Solution: {solution} Topic: {topic}
cpp-qwen35
C / C++ Code Corpus
Description
Teaches domain-specific instruction following and code generation for this expert.
Source
shareAI/CodeChat
Mxode/StackOverflow-QA-C-Language-40k
dumb-dev/cpp-10k
AmareshHebbar/leetcode-codegen-cpp
Formatted for the MoE-orchestrator project
(https://github.com/michaelowusuntim6/MoE-orchestrator). Expert target:
code_cpp.
Format
Each record is a JSON object with a messages field formatted for Qwen3.5's… See the full description on the dataset page: https://huggingface.co/datasets/michaelowusuntim6/cpp-qwen35.dummy-pie-cpp-oracle-seed
Dummy PIE C++ Oracle Seed
This is a tiny 5-row Hugging Face dataset skeleton for generating oracle SFT data for a PIE-style C++ optimization task.
The rows are intentionally synthetic. They define the exact shape we want for real PIE C++ tasks without exposing or depending on private prepared artifacts.
The goal is to give an oracle-data writer a clean contract:
read a C++ optimization task;
produce one verified optimized solution;
return the result in chat/SFT format;
include… See the full description on the dataset page: https://huggingface.co/datasets/TokenBender/dummy-pie-cpp-oracle-seed.CPPB
CPPB
Summary
CPPB is the public release surface for the Controlled Prompt-Privacy Benchmark introduced in BodhiPromptShield: Pre-Inference Prompt Mediation for Suppressing Privacy Propagation in LLM/VLM Agents.
This Hugging Face package intentionally releases the benchmark-authored prompt manifest and template-stratified train/dev/test split, not raw third-party prompts, source images, or end-to-end OCR assets. Each row is a controlled prompt stub with benchmark metadata… See the full description on the dataset page: https://huggingface.co/datasets/mabo1215/CPPB.glm47-aider-cpp-rl-tasks
GLM-4.7 Aider-style C++ RL tasks v2
This access-controlled package contains 253 independently authored C++ repository-editing tasks used for executable-reward post-training of GLM-4.7-Flash.
These are training tasks, not official Aider or Exercism benchmark tasks. The official 26-task Aider C++ evaluation set is not included and remains evaluation-only. The package proves zero official task-ID overlap.
Each task provides an Aider whole-file-edit prompt, starter repository… See the full description on the dataset page: https://huggingface.co/datasets/TokenBender/glm47-aider-cpp-rl-tasks.llama-cpp-cuda12-ada89llama.cpp CUDA 12.8.1, target GPU: sm89
If you're new or just starting to learn setting up your own inferences this llama.cpp wheel will work for if you're using python, CUDA 12.8.1 with one of the following Ada Lovelace generation (sm_89) GPU's:
NVIDIA L4
NVIDIA L40
NVIDIA L40S
NVIDIA RTX 6000 Ada Generation
NVIDIA RTX 5000 Ada Generation
NVIDIA RTX 4500 Ada Generation
NVIDIA RTX 4000 Ada Generation
NVIDIA RTX 4000 SFF Ada Generation
NVIDIA RTX 2000 Ada Generation
NVIDIA GeForce RTX 4090… See the full description on the dataset page: https://huggingface.co/datasets/juiceb0xc0de/llama-cpp-cuda12-ada89.cpp-compiler-prefs
C++ compiler preferences (DPO)
Offline preferences: chosen answers compile; rejected answers fail g++.
cpp-compiler-curriculum
C++ compiler curriculum (SFT)
Synthetic C++ examples derived from Ubuntu toolchain headers and verified with g++ -std=c++20.
cpp-compiler-grpo
C++ compiler GRPO prompts
Prompt-only dataset for GRPO with g++ compile/run rewards.