AI Paper Atlas

Research map / Benchmarks / WikiText

WikiText leaderboard

20 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on WikiText. Best perplexity: 5.55 (FP8 (2), ICLR 2025).

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 · perplexity

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

#perplexity ↓MethodPaperTopic
15.55FP8 (2)
Llama2 7B
Scaling FP8 training to trillion-token LLMs
ICLR 2025 · Zero-shot accuracy and perplex
Quantization
26.61ProxSparse
Llama-2-13b
PROXSPARSE: REGULARIZED LEARNING OF SEMI-STRUCTURED SPARSITY MASKS FOR PRETRAINED LLMS
ICML 2025 · Table 1
Language, safety and interpretability
36.673TEAL 50%
Llama-3-8B
Training-Free Activation Sparsity in Large Language Models
ICLR 2025 · Comparison to 2:4 Weight Spars
Quantization
47.86ARM
Llama-3.1-8B
Attention with Routed-Memory for Learnable Sparse Control
ICML 2026 · Table 1
Attention and transformers
58.3KDP (Ours)
LLaMA-2-7B
KDP: Simplifying Representation Dynamics in Kernel Space
ICLR 2026 · Table 2
RNNs and normalization
68.649DFloat11
Llama 3.1 8B Instruct
70% Size, 100% Accuracy: Lossless LLM Compression for Efficient GPU Inference via Dynamic-Length Float (DFloat11)
NeurIPS 2025 · Accuracy and perplexity table
Quantization
711.1Fixing LLaMA2 13B w/ TempNet
LLaMA2 13B
To Cool or not to Cool? Temperature Network Meets Large Foundation Models via DRO
ICML 2024 · Table 2
Language, safety and interpretability
812.46CAAT-Net
Llama2-7B
Tensor-Parallelism with Partially Synchronized Activations
NeurIPS 2025 · Zero-shot accuracy table
Reasoning and chain of thought
917.35PaTH-FoX
760M Transformer
PaTH Attention: Position Encoding via Accumulating Householder Transformations
NeurIPS 2025 · Table 1
Attention and transformers
1017.87Selective RoPE
GLA 1.3B
Selective Rotary Position Embedding
ICLR 2026 · Table 1
Attention and transformers
10 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.