RULER shows most long-context LMs drop sharply in performance on complex tasks as length and difficulty increase, with only half maintaining results at 32K tokens.
hub Mixed citations
DeepSpeed Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer Models
Mixed citation behavior. Most common role is background (56%).
56
Pith papers citing it
Background
56% of classified citations
abstract
Computation in a typical Transformer-based large language model (LLM) can be characterized by batch size, hidden dimension, number of layers, and sequence length. Until now, system works for accelerating LLM training have focused on the first three dimensions: data parallelism for batch size, tensor parallelism for hidden size and pipeline parallelism for model depth or layers. These widely studied forms of parallelism are not targeted or optimized for long sequence Transformer models. Given practical application needs for long sequence LLM, renewed attentions are being drawn to sequence parallelism. However, existing works in sequence parallelism are constrained by memory-communication inefficiency, limiting their scalability to long sequence large models. In this work, we introduce DeepSpeed-Ulysses, a novel, portable and effective methodology for enabling highly efficient and scalable LLM training with extremely long sequence length. DeepSpeed-Ulysses at its core partitions input data along the sequence dimension and employs an efficient all-to-all collective communication for attention computation. Theoretical communication analysis shows that whereas other methods incur communication overhead as sequence length increases, DeepSpeed-Ulysses maintains constant communication volume when sequence length and compute devices are increased proportionally. Furthermore, experimental evaluations show that DeepSpeed-Ulysses trains 2.5x faster with 4x longer sequence length than the existing method SOTA baseline.
hub tools
citation-role summary
background 10
method 8
citation-polarity summary
representative citing papers
Presents a byte-native LLM with bespoke tokenizer achieving 69-98% accuracy on malware family and architecture classification from raw bytes.
LongLive-2.0 delivers an NVFP4 parallel infrastructure that enables direct training of long multi-shot autoregressive diffusion video models and achieves up to 2.15x training and 1.84x inference speedups on Blackwell and other GPUs.
CausalCine enables real-time causal autoregressive multi-shot video generation via multi-shot training, content-aware memory routing for coherence, and distillation to few-step inference.
Autopoiesis uses LLM-driven program synthesis to evolve serving policies online during deployment, delivering up to 53% and average 34% gains over prior LLM serving systems under runtime dynamics.
KV cache compression causes task-dependent degradation in high-density reasoning due to disrupted CoT links; ShotKV mitigates this by preserving few-shot examples as indivisible semantic units through phase separation, delivering 9-18% accuracy gains and 11% latency reduction.
DuoAttention identifies retrieval heads requiring full KV cache and streaming heads using constant-length cache to reduce memory and latency in long-context LLM inference.
Ring Attention uses blockwise computation and ring communication to let Transformers process sequences up to device-count times longer than prior memory-efficient methods.
A CPU-GPU hybrid design with stream-loading prefill, expert parallelism, and disaggregation achieves cloud SLOs for local MoE inference on dual-socket CPUs and consumer GPUs.
FlashCP introduces Whole-Doc sharding, sharding-aware KV communication, and a heuristic for mixed sharding plans, claiming up to 1.63x speedup over prior CP methods for LLM training.
AAD-1 uses a causal generator with a bidirectional holistic discriminator plus phased distribution matching before adversarial training to reach state-of-the-art one-step autoregressive video generation on VBench.
Schedule-level shared-prefix reuse decouples prefix and suffix passes in GRPO training to compute shared prefixes once, delivering up to 4.395x speedup and 59.1% HBM reduction while preserving numerical equivalence.
LVSA is a training-free block-sparse attention technique combining structured windows with rotating global anchors that reduces inference compute 2.98-3.33x on video diffusion models at extended horizons while remaining quality-neutral or positive.
DualKV eliminates redundant prompt replication in RL training attention kernels via fused dual-KV CUDA operations and token repacking, delivering 1.63-3.82x policy-update speedups while remaining mathematically equivalent to standard attention.
ChunkFlow achieves up to 1.28x step-time speedup and up to 49% lower peak GPU memory for DiT inference by using a first-order model to guide communication-aware chunked prefetching.
MegaScale-Omni delivers 1.27x-7.57x higher throughput for dynamic multimodal LLM training by decoupling encoder and LLM parallelism, using unified colocation, and applying adaptive workload balancing.
Priming transfers knowledge from pre-trained Transformers to hybrid SSM-attention models, recovering performance with minimal additional tokens and showing Gated KalmaNet outperforming Mamba-2 on long-context reasoning at 32B scale.
SPIN co-designs sparse attention with hierarchical memory to achieve 1.66-5.66x higher throughput, 7-9x lower TTFT, and up to 58% lower TPOT than vLLM and original sparse implementations.
CommFuse eliminates tail latency in communication-computation overlap for distributed LLM training by decomposing collective operations into P2P communications and fusing them with fine-grained computation scheduling.
Hive is a multi-agent infrastructure with a logits cache for reducing cross-path redundancy in sampling and agent-aware scheduling for better compute and KV-cache allocation, shown to deliver 1.11x-1.76x speedups and 33%-51% lower hotspot miss rates.
CoCoDiff achieves 3.6x average and 8.4x peak speedup for distributed DiT inference on up to 96 GPU tiles via tile-aware all-to-all, V-first scheduling, and selective V communication.
LingBot-Map is a streaming 3D reconstruction model built on a geometric context transformer that combines anchor context, pose-reference window, and trajectory memory to deliver accurate, drift-resistant results at 20 FPS over sequences longer than 10,000 frames.
OmniShow unifies text, image, audio, and pose conditions into an end-to-end model for high-quality human-object interaction video generation and introduces the HOIVG-Bench benchmark, claiming state-of-the-art results.
LPM 1.0 generates infinite-length, identity-stable, real-time audio-visual conversational performances for single characters using a distilled causal diffusion transformer and a new benchmark.
citing papers explorer
-
CausalCine: Real-Time Autoregressive Generation for Multi-Shot Video Narratives
CausalCine enables real-time causal autoregressive multi-shot video generation via multi-shot training, content-aware memory routing for coherence, and distillation to few-step inference.
-
Autopoiesis: A Self-Evolving System Paradigm for LLM Serving Under Runtime Dynamics
Autopoiesis uses LLM-driven program synthesis to evolve serving policies online during deployment, delivering up to 53% and average 34% gains over prior LLM serving systems under runtime dynamics.
-
ChunkFlow: Communication-Aware Chunked Prefetching for Layerwise Offloading in Distributed Diffusion Transformer Inference
ChunkFlow achieves up to 1.28x step-time speedup and up to 49% lower peak GPU memory for DiT inference by using a first-order model to guide communication-aware chunked prefetching.
-
MegaScale-Omni: A Hyper-Scale, Workload-Resilient System for MultiModal LLM Training in Production
MegaScale-Omni delivers 1.27x-7.57x higher throughput for dynamic multimodal LLM training by decoupling encoder and LLM parallelism, using unified colocation, and applying adaptive workload balancing.
-
OmniShow: Unifying Multimodal Conditions for Human-Object Interaction Video Generation
OmniShow unifies text, image, audio, and pose conditions into an end-to-end model for high-quality human-object interaction video generation and introduces the HOIVG-Bench benchmark, claiming state-of-the-art results.
-
LPM 1.0: Video-based Character Performance Model
LPM 1.0 generates infinite-length, identity-stable, real-time audio-visual conversational performances for single characters using a distilled causal diffusion transformer and a new benchmark.
-
An Efficient Hybrid Sparse Attention with CPU-GPU Parallelism for Long-Context Inference
Fluxion achieves 1.5x-3.7x speedup in long-context LLM inference with CPU KV caches while limiting accuracy degradation to at most 0.26 relative to full attention.
-
Seedance 1.0: Exploring the Boundaries of Video Generation Models
Seedance 1.0 generates 5-second 1080p videos in about 41 seconds with claimed superior motion quality, prompt adherence, and multi-shot consistency compared to prior models.