AI Paper Atlas

Research map / Benchmarks / LAMBADA

LAMBADA leaderboard

41 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on LAMBADA. Best accuracy: 88.52 (Non-uniform, AAAI 2024).

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.

LAMBADA · accuracy

29 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
188.52Non-uniform
LLaMA-7B
What Makes Quantization for Large Language Model Hard? An Empirical Study from the Lens of Perturbation
AAAI 2024 · Table: Comparison of uniform a
Language, safety and interpretability
286.4GPT-3 Few-Shot
GPT-3 175B
Language Models are Few-Shot Learners
NeurIPS 2020 · Table 3
Instruction and fine-tuning
379.2354Norm-Tweaking W4
LLaMa-65B
Norm Tweaking: High-Performance Low-Bit Quantization of Large Language Models
AAAI 2024 · Table: LAMBADA results
Quantization
470.61.3B+PR-MoE-64/128
1.3B
DeepSpeed-MoE: Advancing Mixture-of-Experts Inference and Training to Power Next-Generation AI Scale
ICML 2022 · Table: Zero-shot evaluation re
Mixture of experts
566.3CaDDi-AR
QWen2-1.5B
Non-Markovian Discrete Diffusion with Causal Language Models
NeurIPS 2025 · Table: reasoning benchmarks
Fast diffusion sampling
663.81.3B-ADO
1.3B
Adaptive Data Optimization: Dynamic Sample Selection with Scaling Laws
ICLR 2025 · Table 1
RNNs and normalization
763.71BiDAF-SA-Early + ELMo + CorefAll
BiDAF-SA-Early
Attending to Entities for Better Text Understanding
AAAI 2020 · Main results table
Attention and transformers
862.1FP8 (2)
Llama2 7B
Scaling FP8 training to trillion-token LLMs
ICLR 2025 · Zero-shot accuracy and perplex
Quantization
957MoE++ 7B/(16+4)EMoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts
ICLR 2025 · Table 1
Mixture of experts
1054.34RSEResidual Shuffle-Exchange Networks for Fast Processing of Long Sequences
AAAI 2021 · Table: Accuracy on LAMBADA wor
Attention and transformers
19 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

LAMBADA · perplexity

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

#perplexity ↓MethodPaperTopic
12.97Fixing 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
24.87MuonSphere
Dense 1.7B (Qwen3-1.7B architecture)
Controlled LLM Training on Spectral Sphere
ICML 2026 · Table 3
RNNs and normalization
35.9FP8 (2)
Llama2 7B
Scaling FP8 training to trillion-token LLMs
ICLR 2025 · Zero-shot accuracy and perplex
Quantization
48.5Selective RoPE
GLA 1.3B
Selective Rotary Position Embedding
ICLR 2026 · Table 1
Attention and transformers
512.2Block Transformer
Pythia
Block Transformer: Global-to-Local Language Modeling for Fast Inference
NeurIPS 2024 · Table 1
Attention and transformers
612.51w/ dynamic conv. low-rank (all linear)
Transformer 1.82B (100B tokens)
Dynamic Convolutions Improve Transformers
NeurIPS 2026 · Table 1
Attention and transformers
715.19Q-Delta
1.3B, 30B tokens
Q-Delta: Beyond Key–Value Associative State Evolution
ICML 2026 · Table 2
Attention and transformers
815.9MambaTreeL
Mamba 130M
MambaTree: Tree Topology is All You Need in State Space Model
NeurIPS 2024 · Table 3
CNN features
916.23PaTH-FoX
760M Transformer
PaTH Attention: Position Encoding via Accumulating Householder Transformations
NeurIPS 2025 · Table 1
Attention and transformers
1018.09Gated DeltaNet w/ Log-Linear
LM pretraining 50B tokens, 16K context
Log-Linear Attention
ICLR 2026 · Table 3
Attention and transformers
6 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.