Skip to content
AI NEWS.ccAI正在发生 未来正在更新
Bookmarks
THE JOURNAL
AI NEWS JOURNAL
AI, IN PERSPECTIVE

Search AI news

Search headlines, summaries and articles to connect the dots.

Newest first

arXiv cs.DC

ReKV: Overlapping Quantized KV Transfer and Suffix Recomputation

arXiv:2611.08770v1 Announce Type: new Abstract: Prefill--decode disaggregation moves a prompt's key--value (KV) cache between workers before the first token can be generated. Reducing this handoff to either cache compression or recomputation misses a useful choice: which layers to transmit, at what precision, and which layers to execute again on the decode worker. ReKV transfers a quantized prefix KV cache and one BF16 boundary state, then reconstructs a contiguous suffix while the prefix is in flight. In a model-shaped synthetic benchmark on cross-node H100s, the Q8 configuration meeting separately evaluated quality criteria reduces 8K handoff latency by $2.36\times$, $2.84\times$, and $3.65\times$ over BF16 transfer for Qwen2.5-7B, Qwen2.5-3B, and Qwen3-8B. It is also faster than the best measured transfer-only or recompute-only action at those points. A separate real-weight BF16 recomputation experiment reduces prefill-to-token-return time by $1.16$--$1.63\times$. The preferred action changes with prompt length and link bandwidth.

llama.cpp releases

b12191

The latest update to llama.cpp introduces enhanced MoE (Mixture of Experts) cache functionality, enabling computation of cache misses on the CPU with a configurable CPU share. A new tuning script, moe-cache-tune.py, automatically determines optimal cache size and CPU share settings based on system performance. The update also improves scheduling efficiency, removes redundant sync operations, and refactors code for better maintainability, while maintaining compatibility across multiple platforms including macOS, Linux, Windows, and Android.

llama.cpp releases

b12208

The ggml library has been updated to make tied selections in the radix Top-K algorithm deterministic on CUDA and Metal backends by replacing atomic compaction loops with a deterministic scheme using simdgroup scans and coalesced memory accesses. This change ensures byte-identical outputs across runs, improves performance by 20-45% on M5 Ultra hardware, and enables deterministic testing for Metal and ROCm. Redundant CUDA kernel resets were removed, and Vulkan/Metal changes were reduced or skipped.

llama.cpp releases

b12179

The update introduces support for 3D convolution (conv_3d) on Hexagon hardware by generalizing the existing conv_2d implementation. It leverages the HMX tiled path, folds kernel dimensions into a virtual row, and uses a shared solver with an is3d flag. The host computes unfold divisors to avoid device-side division, prevents DMA queue overflow, and improves tile size selection based on VTCM usage. Additional changes include handling arbitrary positive strides, preserving load/shuffle paths for small strides, and fixing compiler warnings.

llama.cpp releases

b12180

The release adds the Kimi-K3.1-Pro text model to llama.cpp, replacing full-attention MLA layers in K3 with DSA (Dynamic Sparse Attention) layers featuring a lightning indexer. The indexer uses an additive mask over the cache and reuses sparse flash attention paths, while preserving existing KDA layers and MoE components. The model's architecture includes four indexer tensors per DSA layer, hybrid memory for caching, and maintains quantization integrity by excluding indexer parameters from quant overrides. Verified against HF checkpoint with identical PPL beyond context length, and supports multiple platforms including macOS, Linux, Windows, Android, and iOS.

llama.cpp releases

b12189

The metal backend's kernel_group_norm_f32 was updated to correctly handle tensors with ne03 > 1 and non-contiguous inputs by dispatching one threadgroup per (group, i03) pair, computing per-batch statistics, clamping group element count for incomplete groups, and supporting strided access via separate strides for source and destination. The fix also restricts GROUP_NORM to f32 inputs and improves performance through specialized pipeline variants.

llama.cpp releases

b12143

The Vulkan implementation of the Fast Walsh-Hadamard Transform (FWHT) has been enhanced to support block widths up to 8192, using shared-memory shaders for larger blocks, and now includes F16 input capability via shader-float16, with pipelines created only when device limits allow. This update aligns Vulkan with CUDA's capabilities and improves efficiency by eliminating the need for F32-to-F16 conversions.

arXiv cs.LG

LowRAM: Low-Rank Keys and Amortised Values for Composable KV Cache Compression at Test-Time

arXiv:2611.07025v1 Announce Type: new Abstract: Autoregressive decoding relies on the KV cache, whose memory footprint grows linearly with context length and caps batch size and throughput. Existing post-training methods exploit redundancy across tokens, layers, and feature dimensions, or lower numerical precision. We identify a complementary axis: within each attention head, values are largely predictable from their corresponding keys. We exploit this with value cache amortisation, which fits a lightweight per-head predictor at test time and stores residuals only for the tokens it reconstructs worst, giving $4\times$ near-lossless value compression. The predictor can be initialised by meta-learning for substantially reduced prefill overhead, or fitted from scratch in settings that allow a one-off cost, such as multi-query inference. We pair value amortisation with cross-layer low-rank key compression to produce LowRAM, an asymmetric compression pipeline that stays close to lossless across three model families. By reconstructing only a selected subset of KV pairs at each decoding step, LowRAM can keep decoding overhead minimal and achieve up to $5\times$ higher peak decode throughput at $4\times$ compression. LowRAM further composes with quantisation and eviction, reaching $26.9\times$ compression at $94.5\%$ of baseline performance on long contexts. These results show that treating keys and values as structurally distinct enables efficient, aggressive, and composable KV cache compression.

vLLM · 官方技术博客

Migrating vLLM to the LibTorch Stable ABI

How vLLM migrated its CUDA build and compiled dependencies to the LibTorch Stable ABI, enabling compatible PyTorch upgrades without rebuilding vLLM.

arXiv cs.DC

Not All Loops Cost the Same: Phase-Dependent Depth for Serving Looped Transformers

arXiv:2611.07211v1 Announce Type: new Abstract: Looped transformers scale inference-time reasoning by repeatedly reusing a shared core, increasing computation without increasing parameter count. Serving prefilling and decoding with a uniform loop budget couples their costs. We present LoopDuet, a phase-dependent depth and training-free serving approach that selects the two budgets jointly while preserving decoding depth at prefilling boundaries. Our vLLM implementation gathers tokens requiring further recurrence into compact batches, turning the selected allocation into executable savings with existing weights and attention kernels. At runtime, a queue-age controller adjusts prefilling depth to retain quality as demand changes. Experiments on the full HotpotQA dataset with one A100-80GB demonstrate two benefits of this flexibility. First, separate phase budgets improve the quality--throughput trade-off: Huginn delivers 19\% more throughput than the fastest tested uniform setting meeting the same accuracy criterion, while a higher-throughput operating point gains 6.04 EM points over uniform depth eight at 97\% of its throughput. Second, the same execution mechanism retains more quality under load: on two evaluated OLMo arrival traces, adaptive prefilling gains 0.92--1.35 EM points over the most accurate tested fixed-depth controls at the same 100\% attainment of a one-second time-to-first-token target. Together, these results establish LoopDuet as a practical way to turn flexible model computation into higher serving capacity and better quality under load.

arXiv cs.CL

Where Should a KV-Cache Repair Budget Go? Distractor Structure and Contextual Compatibility in Non-Prefix Reuse

arXiv:2611.06397v1 Announce Type: new Abstract: Non-prefix KV-cache reuse encodes retrieved chunks independently and concatenates their caches. To find where a small repair budget should be spent, we vary the chunks around the answer-bearing one (the target) across three 7B-8B models and four datasets of dialogue memory, open-domain QA and API documentation. Reuse loses 19-100% of the context-dependent questions that full prefill answers, and it loses more when the other chunks are of the same kind as the target. Because the recoverable damage is spread over these other chunks, restoring a few of them recovers little, and the budget is better spent on the target. This works through contextual compatibility: recomputing the target against the reused cache beats inserting its full-prefill KV in all seven settings tested, although the recomputed KV stays far from the full-prefill KV. Question attention over reused caches, however, selects the target in at most 18.8% of development questions. A retriever finds it more often, and recomputing its top chunk after moving it next to the question significantly beats InfoFlow-style reordering at the same 5% budget in six of nine open-domain QA and API-documentation settings. On dialogue memory, though, recomputing chunks never significantly beats the token-level selector chosen for each setting.

arXiv cs.CV

From Tokens to Trajectories: Parameter-Efficient Continuous Adaptation of Unified Multimodal Models

arXiv:2611.05451v1 Announce Type: new Abstract: While discrete tokenization in Unified Multimodal Models (UMMs) establishes strong semantic reasoning, it introduces a strict quantization bottleneck that degrades fine-grained visual details, often resulting in structurally accurate but artifact-heavy images. We introduce FlowShift, a parameter-efficient post-training framework that bypasses this bottleneck by converting pre-trained discrete UMMs into continuous rectified-flow generators. Instead of discarding the backbone's semantic priors, FlowShift replaces discrete visual interfaces with continuous latent projections and applies LoRA to the attention layers. To efficiently boost visual quality without semantic drift, we propose Dual-Objective Velocity Vector Distillation (DOVD), which aligns the model with a high-fidelity continuous teacher and ground-truth continuous trajectories. This approach drastically improves perceptual quality while preserving prompt adherence and reasoning. When applied to leading discrete UMMs, FlowShift significantly improves COCO-30K FID and aesthetic scores, while retaining over 99.5% of multimodal understanding capabilities. Ablations confirm DOVD drives generation quality. Ultimately, FlowShift achieves high-fidelity continuous generation without sacrificing pre-trained semantic grounding.

llama.cpp releases

b12094

Optimized TOP_K performance on Hexagon backend by skipping bitonic sort for tiles whose maximum cannot surpass the k-th best value, replacing tile sort with top_k_scan() to reduce 256-float blocks using HVX max-reduce, and inserting only surviving elements into a sorted k-entry list. Supports k values from 2 to 64 and row widths over 1024 in both single-row and multi-row paths. NaN treated as -INF, ties preserve first-seen index. Eliminated scalar reads from VTCM in hex-top-k.

llama.cpp releases

b12102

The ggml library now checks view bounds based on the actual bytes reached by strides, improving accuracy for strided views. Views are required to start from contiguous elements along dimension 0, ensuring correct data access. Convolution windows in delta_net are created with all rows at once to prevent stride-related issues.

arXiv cs.DC

LiquidCP: Decoupling Attention Tasks from Communication and Kernel Boundaries in Context Parallelism

arXiv:2611.04401v1 Announce Type: new Abstract: Context parallelism (CP) partitions long sequences across GPUs for LLM training and inference. For variable-length training and full prefill, attention-balanced placement can leave token loads uneven, while ring execution repeats overhead at each computation and communication step. In stage-serialized decode CP (DCP), long-running tasks delay dependent work and leave substantial GPU hardware idle. We propose LiquidCP, which unifies training and inference through a common model of Transfer, Compute, and Reduction tasks with tile-level dependencies, decoupling attention tasks from communication and kernel boundaries. For training and full prefill, MegaRing incorporates BuddyRing placement to balance attention and token loads, and coalesces ready KV regions during forward attention to amortize overhead. For decoding and mixed batches, MegaDCP selects history-KV splits by jointly modeling the completion of attention and its dependent reductions. A shared persistent GPU-resident engine executes both designs. Compared with the fastest baseline per configuration on eight H100 GPUs, MegaRing achieves geometric-mean core-attention speedups of $1.23\times$ forward and $1.18\times$ backward, and accelerates a Transformer layer by $1.08$--$1.16\times$. MegaDCP achieves geometric-mean core-attention speedups of $1.20\times$ for decode-only and $1.35\times$ for mixed batches. In vLLM serving with KV prefix cache, it achieves an average time-to-first-token speedup of $1.22\times$.

vLLM · 官方技术博客

vLLM Delivers Day-0 Support for Kimi K3.1 Pro

vLLM supports Kimi K3.1 Pro on day 0: sparse attention on a KDA hybrid, a packed KV cache with FP8 or NVFP4 latents and MXFP4 index keys, token-sharded prefill, decode context parallelism, and fused low-latency decode.

arXiv cs.DC

Fairway: Load-Aware KV Cache Reuse for LLM Prefill Scheduling

arXiv:2611.02082v1 Announce Type: new Abstract: LLM-based applications often employ agentic workflows and retrieval-augmented generation (RAG), which build long prompts from shared context. When prefixes are shared across requests, serving engines reuse the key-value (KV) cache to skip parts of the prefill computation. Reuse is typically graded: the uncached work of requests in a queue can differ by up to 100x. In this setting, the order of scheduling determines the time-to-first-token (TTFT) for the queued requests. KV reuse-first scheduling, which prioritizes the requests with the longest KV cache hit, gives the lowest median TTFT but starves expensive requests under load spikes; first-come-first-served (FCFS) scheduling bounds the tail but raises the median TTFT. We present Fairway, a prefill scheduler built on adaptive reuse prioritization: it favors KV cache reuse while queues are short, and shifts toward fair allocation as requests wait longer relative to their own prefill time. Fairway employs a virtual dual lane model: a KV reuse-first Fast lane and a Fair lane atop the serving engine's waiting queue that enables speeding up of cheap reuse requests without starving expensive ones. This model automatically adapts to the workload without any per-workload tuning. Fairway lowers the tail TTFT at no cost to the median in transient overload. Across workloads and load levels, Fairway matches or beats the best baseline, with a p99 TTFT up to 8.6x lower than prior prefix-aware schedulers. During load spikes, it keeps the tail of FCFS at a 6x lower median TTFT.

llama.cpp releases

llama.cpp · b12065

kv-cache : never return 0 kv_rows for sequences without cells (#31570) A cache that shares the cells of another one (f.ex. the gemma4 assistant) can build a graph before its sequences have any cells, f.ex. at the server warmup. get_n_kv_rows() then returned 0, which means that kv_rows are not used, so the graph fell back to the dense KQ mask and the scheduler re-planned for it. Later graphs with larger masks then aborted with GGML_SCHED_DEBUG_REALLOC=1. **Website:** - **Attestations:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12065/llama-b12065-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel (x64)](https://github.com/ggml-org/l

llama.cpp releases

b11923

ggml : relax scale to non-contiguous views (#31146) * ggml : relax scale to non-contiguous views SCALE requires src0 to be 1D-contiguous. Relax this to dim-0 contiguity with arbitrary row strides (ggml_is_contiguous_rows), same as the unary ops in the Metal backend. - ggml: relax the assert in ggml_scale_impl - cpu: address src0 rows through their byte strides (f32, bf16) - cuda: stride-aware kernel, fastdiv row decomposition - metal: support bf16 scale (guarded unary instantiations) - sycl: keep the gate at contiguous, the kernel is flat f32 Add view variants to test_scale. Verified on CPU, Metal and CUDA. Assisted-by: dsh:llama.cpp/ggml-org/Qwen3.8-Flash-Next-GGUF (RTX 5090 32GB, DDR5 128GB) * cuda : skip the softcap fusion on non-contiguous views The fused softcap kernel reads src and w

Timeline and related coverage