The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits
Paper
• 2402.17764 • Published
• 627
LM Inference
-
-
BitNet: Scaling 1-bit Transformers for Large Language Models
Paper • 2310.11453 • Published • 106 -
Mixture-of-Depths: Dynamically allocating compute in transformer-based language models
Paper • 2404.02258 • Published • 107 -
TransformerFAM: Feedback attention is working memory
Paper • 2404.09173 • Published • 43 -
Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length
Paper • 2404.08801 • Published • 66 -
Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention
Paper • 2404.07143 • Published • 111 -
Eagle and Finch: RWKV with Matrix-Valued States and Dynamic Recurrence
Paper • 2404.05892 • Published • 40 -
MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding
Paper • 2404.05726 • Published • 23 -
LongRoPE: Extending LLM Context Window Beyond 2 Million Tokens
Paper • 2402.13753 • Published • 116 -
How Good Are Low-bit Quantized LLaMA3 Models? An Empirical Study
Paper • 2404.14047 • Published • 45 -
SnapKV: LLM Knows What You are Looking for Before Generation
Paper • 2404.14469 • Published • 27 -
LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding
Paper • 2404.16710 • Published • 80 -
Octopus v4: Graph of language models
Paper • 2404.19296 • Published • 118 -
Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting
Paper • 2404.18911 • Published • 30 -
LoRA Land: 310 Fine-tuned LLMs that Rival GPT-4, A Technical Report
Paper • 2405.00732 • Published • 122 -
Imp: Highly Capable Large Multimodal Models for Mobile Devices
Paper • 2405.12107 • Published • 29 -
Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality
Paper • 2405.21060 • Published • 68 -
Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling
Paper • 2406.07522 • Published • 40 -
Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B
Paper • 2406.07394 • Published • 29 -
Large Language Monkeys: Scaling Inference Compute with Repeated Sampling
Paper • 2407.21787 • Published • 13 -
ThinK: Thinner Key Cache by Query-Driven Pruning
Paper • 2407.21018 • Published • 32 -
A Comprehensive Evaluation of Quantized Instruction-Tuned Large Language Models: An Experimental Analysis up to 405B
Paper • 2409.11055 • Published • 17 -
Discovering the Gems in Early Layers: Accelerating Long-Context LLMs with 1000x Input Token Reduction
Paper • 2409.17422 • Published • 25 -
Thinking LLMs: General Instruction Following with Thought Generation
Paper • 2410.10630 • Published • 20 -
VPTQ: Extreme Low-bit Vector Post-Training Quantization for Large Language Models
Paper • 2409.17066 • Published • 28 -
Efficiently Serving LLM Reasoning Programs with Certaindex
Paper • 2412.20993 • Published • 36 -
Token-Budget-Aware LLM Reasoning
Paper • 2412.18547 • Published • 46 -
Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM's Reasoning Capability
Paper • 2411.19943 • Published • 62 -
Low-Bit Quantization Favors Undertrained LLMs: Scaling Laws for Quantized LLMs with 100T Training Tokens
Paper • 2411.17691 • Published • 13 -
Star Attention: Efficient LLM Inference over Long Sequences
Paper • 2411.17116 • Published • 53 -
SageAttention2 Technical Report: Accurate 4 Bit Attention for Plug-and-play Inference Acceleration
Paper • 2411.10958 • Published • 57 -
BitNet a4.8: 4-bit Activations for 1-bit LLMs
Paper • 2411.04965 • Published • 69 -
1-bit AI Infra: Part 1.1, Fast and Lossless BitNet b1.58 Inference on CPUs
Paper • 2410.16144 • Published • 5 -
FlatQuant: Flatness Matters for LLM Quantization
Paper • 2410.09426 • Published • 15 -
PrefixQuant: Static Quantization Beats Dynamic through Prefixed Outliers in LLMs
Paper • 2410.05265 • Published • 33 -
Tensor Product Attention Is All You Need
Paper • 2501.06425 • Published • 90 -
Test-Time Preference Optimization: On-the-Fly Alignment via Iterative Textual Feedback
Paper • 2501.12895 • Published • 61 -
Qwen2.5-1M Technical Report
Paper • 2501.15383 • Published • 72 -
Reward-Guided Speculative Decoding for Efficient LLM Reasoning
Paper • 2501.19324 • Published • 39 -
DeepRAG: Thinking to Retrieval Step by Step for Large Language Models
Paper • 2502.01142 • Published • 24 -
QuEST: Stable Training of LLMs with 1-Bit Weights and Activations
Paper • 2502.05003 • Published • 44 -
CMoE: Fast Carving of Mixture-of-Experts for Efficient LLM Inference
Paper • 2502.04416 • Published • 12 -
Matryoshka Quantization
Paper • 2502.06786 • Published • 32 -
Lossless Acceleration of Large Language Models with Hierarchical Drafting based on Temporal Locality in Speculative Decoding
Paper • 2502.05609 • Published • 18 -
TransMLA: Multi-head Latent Attention Is All You Need
Paper • 2502.07864 • Published • 57 -
InfiniteHiP: Extending Language Model Context Up to 3 Million Tokens on a Single GPU
Paper • 2502.08910 • Published • 148 -
Diverse Inference and Verification for Advanced Reasoning
Paper • 2502.09955 • Published • 18 -
Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention
Paper • 2502.11089 • Published • 168 -
LightThinker: Thinking Step-by-Step Compression
Paper • 2502.15589 • Published • 31 -
MoBA: Mixture of Block Attention for Long-Context LLMs
Paper • 2502.13189 • Published • 17 -
SpargeAttn: Accurate Sparse Attention Accelerating Any Model Inference
Paper • 2502.18137 • Published • 60 -
SEAP: Training-free Sparse Expert Activation Pruning Unlock the Brainpower of Large Language Models
Paper • 2503.07605 • Published • 66 -
TTRL: Test-Time Reinforcement Learning
Paper • 2504.16084 • Published • 120 -
φ-Decoding: Adaptive Foresight Sampling for Balanced Inference-Time Exploration and Exploitation
Paper • 2503.13288 • Published • 51 -
Plug-and-Play 1.x-Bit KV Cache Quantization for Video Large Language Models
Paper • 2503.16257 • Published • 27 -
FFN Fusion: Rethinking Sequential Computation in Large Language Models
Paper • 2503.18908 • Published • 19 -
LogQuant: Log-Distributed 2-Bit Quantization of KV Cache with Superior Accuracy Preservation
Paper • 2503.19950 • Published • 12 -
AdaptiVocab: Enhancing LLM Efficiency in Focused Domains through Lightweight Vocabulary Adaptation
Paper • 2503.19693 • Published • 76 -
Efficient Inference for Large Reasoning Models: A Survey
Paper • 2503.23077 • Published • 46 -
A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyond
Paper • 2503.21614 • Published • 43 -
Hogwild! Inference: Parallel LLM Generation via Concurrent Attention
Paper • 2504.06261 • Published • 110 -
C3PO: Critical-Layer, Core-Expert, Collaborative Pathway Optimization for Test-Time Expert Re-Mixing
Paper • 2504.07964 • Published • 62 -
Quantization Hurts Reasoning? An Empirical Study on Quantized Reasoning Models
Paper • 2504.04823 • Published • 31 -
HybriMoE: Hybrid CPU-GPU Scheduling and Cache Management for Efficient MoE Inference
Paper • 2504.05897 • Published • 21 -
Accelerate Parallelizable Reasoning via Parallel Decoding within One Sequence
Paper • 2503.20533 • Published • 12 -
PRIMA.CPP: Speeding Up 70B-Scale LLM Inference on Low-Resource Everyday Home Clusters
Paper • 2504.08791 • Published • 139 -
BitNet b1.58 2B4T Technical Report
Paper • 2504.12285 • Published • 83 -
70% Size, 100% Accuracy: Lossless LLM Compression for Efficient GPU Inference via Dynamic-Length Float
Paper • 2504.11651 • Published • 31 -
Efficient Reasoning Models: A Survey
Paper • 2504.10903 • Published • 21 -
Sleep-time Compute: Beyond Inference Scaling at Test-time
Paper • 2504.13171 • Published • 15 -
SpecReason: Fast and Accurate Inference-Time Compute via Speculative Reasoning
Paper • 2504.07891 • Published • 5 -
EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Models
Paper • 2504.15133 • Published • 26