attn-signs/GPTR-8b-v2

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Apr 16, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The attn-signs/GPTR-8b-v2 is an 8 billion parameter reasoning model developed by Reisen Raumberg (Attention Signs team), adapted for Russian text generation. Built upon the YandexGPT-pretrain and previous GPTR versions, this model has undergone a second stage of reinforcement learning (GRPO) to enhance its general reasoning capabilities. It excels in mathematical tasks and can generate conditional and coherent chain-of-thought responses in both Russian and English.

Loading preview...

GPTR-8b-v2: Enhanced Reasoning for Russian Text Generation

The attn-signs/GPTR-8b-v2 is an 8 billion parameter language model developed by the Attention Signs team, specifically Reisen Raumberg. It represents the second reinforced GRPO version in its lineage, building upon YandexGPT-pretrain and prior GPTR-8b iterations.

Key Capabilities

  • Advanced Reasoning: The model has undergone a second stage of reinforcement learning (RL GRPO) to significantly improve its general reasoning abilities.
  • Mathematical Proficiency: Demonstrates enhanced performance in solving mathematical tasks and proving mathematical statements.
  • Coherent Chain-of-Thought: Capable of generating conditional and coherent chain-of-thought responses.
  • Bilingual Support: Supports text generation in both Russian (RU) and English (EN).
  • Training: Fine-tuned from YandexGPT-5-lite-8B-pretrain using the MyLLM framework developed by Attention Signs, with approximately 24 hours of NVIDIA A100 GPU training.

Good For

  • Applications requiring robust reasoning in Russian.
  • Mathematical problem-solving and proof generation.
  • Generating structured and logical responses based on complex prompts.
  • Developers looking for a specialized model with strong analytical capabilities in a bilingual context.