Llama-3

'LLaMA3' is a benchmark used to evaluate the multi-token prediction capabilities of large language models, demonstrating improvements in acceptance length and token throughput through a training-free probing approach.

Papers using Llama-3 (12)

Efficient Training-Free Multi-Token Prediction via Embedding-Space Probing2026Analyzing the Effects of Supervised Fine-Tuning on Model Knowledge from Token and Parameter Levels2025IntraSlice: Towards High-Performance Structural Pruning with Block-Intra PCA for LLMs2026Compressing LLMs with MoP: Mixture of Pruners2026Leveraging KV Similarity for Online Structured Pruning in LLMs2025NIRVANA: Structured pruning reimagined for large language models compression2025PRIMA.CPP: Speeding Up 70B-Scale LLM Inference on Low-Resource Everyday Home Clusters2025UniQL: Unified Quantization and Low-rank Compression for Adaptive Edge LLMs2025Precision Where It Matters: A Novel Spike Aware Mixed-Precision Quantization Strategy for LLaMA-based Language Models2025Bridging the LLM Accessibility Divide? Performance, Fairness, and Cost of Closed versus Open LLMs for Automated Essay Scoring2025Beyond One-Size-Fits-All Pruning via Evolutionary Metric Search for Large Language Models2025Improving Influence-based Instruction Tuning Data Selection for Balanced Learning of Diverse Capabilities2025