AI Paper Atlas

Research map / Benchmarks / WikiText-2

WikiText-2 leaderboard

70 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on WikiText-2. Best perplexity: 2.86 (NLI, ICLR 2026).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

WikiText-2 · perplexity

70 papers, each with its best reported number for its own method; lower is better.

#perplexity ↓MethodPaperTopic
12.86NLI
Llama3-70B
NLI : Non-uniform Linear Interpolation Approximation of Nonlinear Operations for Efficient LLMs Inference
ICLR 2026 · Table 1
Reasoning and chain of thought
22.93FedLMT
Transformer
FedLMT: Tackling System Heterogeneity of Federated Learning via Low-Rank Model Training with Theoretical Guarantees
ICML 2024 · Table 2
Personalized and heterogeneous FL
32.94LaRoSA 25%
LLaMA3-70B
La RoSA: Enhancing LLM Efficiency via Layerwise Rotated Sparse Activation
ICML 2025 · Perplexity results table
Pruning and efficient tuning
43.37HeteroFL a
Transformer
HeteroFL: Computation and Communication Efficient Federated Learning for Heterogeneous Clients
ICLR 2021 · Table 3
Communication-efficient FL
53.5S-Quant (3.8 bit)
Llama 2 70B
S-Quant: Rethinking Weight Quantization with Seed-Based Generation
ICML 2026 · Table 2
Quantization
63.756AdaCubic
RoBERTa
AdaCubic: An Adaptive Cubic Regularization Optimizer for Deep Learning
NeurIPS 2026 · Table 5
Pruning and efficient tuning
73.85Proposed
Mixtral-8x7B
QoS-Efficient Serving of Multiple Mixture-of-Expert LLMs Using Partial Runtime Reconfiguration
ICML 2025 · Quality of Generated Output ta
Mixture of experts
84.09BESA
LLaMA2-70B
BESA: Pruning Large Language Models with Blockwise Parameter-Efficient Sparsity Allocation
ICLR 2024 · Table 1
Pruning and efficient tuning
94.188LSTM, 24M, depth 1On the State of the Art of Evaluation in Neural Language Models
ICLR 2018 · Table: Wikitext-2 results
RNNs and normalization
104.6SliceGPT 25%
Llama 70B
SliceGPT: Compress Large Language Models by Deleting Rows and Columns
ICLR 2024 · Table 1
Pruning and efficient tuning
60 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.