Dataset · Vikhrmodels
Russian SFT instruction dataset
Vikhrmodels/GrandMaster-PRO-MAX
A large Russian instruction-response dataset for training models to follow commands.
What's inside
Contains instructions paired with assistant responses generated entirely from scratch by GPT-4-Turbo-1106 based on the original prompts. The languages are Russian and English, primarily Russian. Metadata indicates a size of 100,000 to 1 million examples across tabular and text modalities.
What you can do with it
- Fine-tuning a Russian instruct model (SFT)
- Training LoRA adapters for instruction following
- Evaluating Russian response quality after SFT
- Tasks
- Text generation
- Languages
- Russian, English
- Size
- 100K–1M rows
- Downloads this month
- 1.6K
- Likes
- 78
Models trained on it
- Vikhrmodels/Vikhr-Llama3.1-8B-Instruct-R-21-09-24 38.3K
- Vikhrmodels/Vikhr-Nemo-12B-Instruct-R-21-09-24Text generation 36.3K
- VlSav/Vikhr-Nemo-12B-Instruct-R-21-09-24-Q4_K_M-GGUF 8.2K
- Vikhrmodels/Vikhr-Gemma-2B-instruct-GGUFText generation 3.9K
- bartowski/Vikhr-Nemo-12B-Instruct-R-21-09-24-GGUFText generation 3.5K
- mradermacher/Viper-9B-i1-GGUF 3.2K
- mradermacher/Mercury-Acad-1.7B-i1-GGUF 3K
- Vikhrmodels/Vikhr-Qwen-2.5-1.5B-Instruct-GGUF 2.4K
How to load it
from datasets import load_dataset
ds = load_dataset("Vikhrmodels/GrandMaster-PRO-MAX")