DeepSeek-R-1

The 'Deepseek R-1' dataset/benchmark is used to evaluate the efficiency and effectiveness of various attention mechanisms, including Multi-head Latent Attention (MLA) and Group Query Attention (GQA), in large language models.

Papers using DeepSeek-R-1 (8)

Information-Aware KV Cache Compression for Long Reasoning2026TransMLA: Multi-head Latent Attention Is All You Need2025 · 6 citesSWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolution2025 · 2 citesLearning a Continue-Thinking Token for Enhanced Test-Time Scaling2025PRIMA.CPP: Speeding Up 70B-Scale LLM Inference on Low-Resource Everyday Home Clusters2025From Harm to Help: Turning Reasoning In-Context Demos into Assets for Reasoning LMs2025RRTL: Red Teaming Reasoning Large Language Models in Tool Learning2025Adaptive Rectification Sampling for Test-Time Compute Scaling2025