arXiv subjects

Huanqi Cao

Publications and source records attributed to Huanqi Cao.

10 recordsLinked to original sources

arXiv2026

DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression

The widespread adoption of long-horizon agents has made model workloads increasingly input-heavy. Although prior work has substantially reduced the cost of long-context computation, prefill remains computationally expensive, and large KV caches continue to strain HBM and SSD capacity and data-transfer bandwidth. Together, these compute, storage, and bandwidth demands constitute the primary bottleneck to further lowering deployment costs. To address this challenge, we introduce DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts (MoE) model with 552B backbone parameters and support for contexts of up to one million tokens. With its Causal Encoder-Decoder (CED) architecture, the model activates 16B parameters per token during decode but only 8B parameters during prefill, substantially improving cost efficiency for agentic workloads. To push the limits of KV cache compression, DeepSeek-V4.1-Flash combines cross-layer KV cache reuse in Compressed Sparse Attention 2 (CSA2) with FP4 KV caching. These designs reduce its global KV cache footprint (always in HBM) to 890 bytes per token, roughly 1/4 of the corresponding footprint of DeepSeek-V4-Flash. Further, through a dedicated deployment optimization known as SWA Bounded Replay, DeepSeek-V4.1-Flash reduces its persistent KV cache footprint (always on SSD or in host memory) to roughly 1/8 of that of DeepSeek-V4-Flash. Despite its much smaller KV cache footprint, the model delivers substantially better performance than the baseline. In addition, we streamline the DeepSeek-V4 architecture and introduce several efficient architectural extensions. We pretrain DeepSeek-V4.1-Flash on a multimodal corpus comprising 45T tokens and conduct comprehensive post-training, yielding strong performance across diverse text-based and multimodal agentic scenarios. Model checkpoints are available at https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash.

cs.CL↗

arXiv2026

DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

Speculative decoding accelerates Large Language Model (LLM) inference by decoupling draft generation from target verification. While recent parallel drafters efficiently propose long token sequences in a single forward pass, they suffer from rapid acceptance decay due to a lack of inter-token dependencies. Furthermore, indiscriminately verifying these extended blocks wastes critical batch capacity on tokens with high rejection risks, severely degrading throughput in high-concurrency serving systems. We introduce DSpark, a speculative decoding framework that unifies high-throughput parallel generation with adaptive, load-aware verification. To maintain draft quality, DSpark utilizes a semi-autoregressive architecture, coupling a parallel backbone with a lightweight sequential module, to introduce intra-block dependency modeling and mitigate suffix decay. To optimize system efficiency, DSpark employs confidence-scheduled verification, dynamically tailoring the verification length for each request based on estimated prefix survival probabilities and engine-specific throughput profiles. On offline benchmarks across diverse domains, DSpark substantially improves the accepted length over state-of-the-art autoregressive and parallel drafters. When deployed within the DeepSeek-V4 serving system under live user traffic, DSpark successfully mitigates verification waste. Compared to the established production baseline (MTP-1), DSpark accelerates per-user generation speeds by 60 to 85 percent at matched throughput levels. More importantly, by preventing severe throughput degradation under strict interactivity constraints, it enables performance tiers that were previously unattainable, shifting the Pareto frontier of our serving system.

cs.AI↗

arXiv2026

DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence

We present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models -- DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) -- both supporting a context length of one million tokens. DeepSeek-V4 series incorporate several key upgrades in architecture and optimization: (1) a hybrid attention architecture that combines Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to improve long-context efficiency; (2) Manifold-Constrained Hyper-Connections (mHC) that enhance conventional residual connections; (3) and the Muon optimizer for faster convergence and greater training stability. We pre-train both models on more than 32T diverse and high-quality tokens, followed by a comprehensive post-training pipeline that unlocks and further enhances their capabilities. DeepSeek-V4-Pro-Max, the maximum reasoning effort mode of DeepSeek-V4-Pro, redefines the state-of-the-art for open models, outperforming its predecessors in core tasks. Meanwhile, DeepSeek-V4 series are highly efficient in long-context scenarios. In the one-million-token context setting, DeepSeek-V4-Pro requires only 27% of single-token inference FLOPs and 10% of KV cache compared with DeepSeek-V3.2. This enables us to routinely support one-million-token contexts, thereby making long-horizon tasks and further test-time scaling more feasible. The model checkpoints are available at https://huggingface.co/collections/deepseek-ai/deepseek-v4.

cs.CL↗

arXiv2026

mHC: Manifold-Constrained Hyper-Connections

Recently, studies exemplified by Hyper-Connections (HC) have extended the ubiquitous residual connection paradigm established over the past decade by expanding the residual stream width and diversifying connectivity patterns. While yielding substantial performance gains, this diversification fundamentally compromises the identity mapping property intrinsic to the residual connection, which causes severe training instability and restricted scalability, and additionally incurs notable memory access overhead. To address these challenges, we propose Manifold-Constrained Hyper-Connections (mHC), a general framework that projects the residual connection space of HC onto a specific manifold to restore the identity mapping property, while incorporating rigorous infrastructure optimization to ensure efficiency. Empirical experiments demonstrate that mHC is effective for training at scale, offering tangible performance improvements and superior scalability. We anticipate that mHC, as a flexible and practical extension of HC, will contribute to a deeper understanding of topological architecture design and suggest promising directions for the evolution of foundational models.

cs.CL↗

arXiv2025

DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models

We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. The key technical breakthroughs of DeepSeek-V3.2 are as follows: (1) DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 and exhibits reasoning proficiency on par with Gemini-3.0-Pro, achieving gold-medal performance in both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI). (3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.

cs.CL↗

arXiv2023

RWKV: Reinventing RNNs for the Transformer Era

Transformers have revolutionized almost all natural language processing (NLP) tasks but suffer from memory and computational complexity that scales quadratically with sequence length. In contrast, recurrent neural networks (RNNs) exhibit linear scaling in memory and computational requirements but struggle to match the same performance as Transformers due to limitations in parallelization and scalability. We propose a novel model architecture, Receptance Weighted Key Value (RWKV), that combines the efficient parallelizable training of transformers with the efficient inference of RNNs. Our approach leverages a linear attention mechanism and allows us to formulate the model as either a Transformer or an RNN, thus parallelizing computations during training and maintains constant computational and memory complexity during inference. We scale our models as large as 14 billion parameters, by far the largest dense RNN ever trained, and find RWKV performs on par with similarly sized Transformers, suggesting future work can leverage this architecture to create more efficient models. This work presents a significant step towards reconciling trade-offs between computational efficiency and model performance in sequence processing tasks.

cs.CL↗

arXiv2023

Mat2Stencil: A Modular Matrix-Based DSL for Explicit and Implicit Matrix-Free PDE Solvers on Structured Grid

Partial differential equation (PDE) solvers are extensively utilized across numerous scientific and engineering fields. However, achieving high performance and scalability often necessitates intricate and low-level programming, particularly when leveraging deterministic sparsity patterns in structured grids. In this paper, we propose an innovative domain-specific language (DSL), Mat2Stencil, with its compiler, for PDE solvers on structured grids. Mat2Stencil introduces a structured sparse matrix abstraction, facilitating modular, flexible, and easy-to-use expression of solvers across a broad spectrum, encompassing components such as Jacobi or Gauss-Seidel preconditioners, incomplete LU or Cholesky decompositions, and multigrid methods built upon them. Our DSL compiler subsequently generates matrix-free code consisting of generalized stencils through multi-stage programming. The code allows spatial loop-carried dependence in the form of quasi-affine loops, in addition to the Jacobi-style stencil's embarrassingly parallel on spatial dimensions. We further propose a novel automatic parallelization technique for the spatially dependent loops, which offers a compile-time deterministic task partitioning for threading, calculates necessary inter-thread synchronization automatically, and generates an efficient multi-threaded implementation with fine-grained synchronization. Implementing 4 benchmarking programs, 3 of them being the pseudo-applications in NAS Parallel Benchmarks with $6.3\%$ lines of code and 1 being matrix-free High Performance Conjugate Gradients with $16.4\%$ lines of code, we achieve up to $1.67\times$ and on average $1.03\times$ performance compared to manual implementations.

cs.DC↗

arXiv2023

PowerFusion: A Tensor Compiler with Explicit Data Movement Description and Instruction-level Graph IR

Deep neural networks (DNNs) are of critical use in different domains. To accelerate DNN computation, tensor compilers are proposed to generate efficient code on different domain-specific accelerators. Existing tensor compilers mainly focus on optimizing computation efficiency. However, memory access is becoming a key performance bottleneck because the computational performance of accelerators is increasing much faster than memory performance. The lack of direct description of memory access and data dependence in current tensor compilers' intermediate representation (IR) brings significant challenges to generate memory-efficient code. In this paper, we propose IntelliGen, a tensor compiler that can generate high-performance code for memory-intensive operators by considering both computation and data movement optimizations. IntelliGen represent a DNN program using GIR, which includes primitives indicating its computation, data movement, and parallel strategies. This information will be further composed as an instruction-level dataflow graph to perform holistic optimizations by searching different memory access patterns and computation operations, and generating memory-efficient code on different hardware. We evaluate IntelliGen on NVIDIA GPU, AMD GPU, and Cambricon MLU, showing speedup up to 1.97x, 2.93x, and 16.91x(1.28x, 1.23x, and 2.31x on average), respectively, compared to current most performant frameworks.

cs.LG↗

arXiv2022

Design and Implementation of ShenWei Universal C/C++

The ShenWei many-core series processors powering multiple cutting-edge supercomputers are equipped with their unique on-chip heterogeneous architecture. They have long required programmers to write separate codes for the control part on Management Processing Element (MPE) and accelerated part on Compute Processing Element (CPE), which is similar to open standards like OpenCL. Such a programming model results in shattered code and bad maintainability, and also make it hard to migrate existing projects targeting commodity processors. Borrowing the experience from CUDA and DPC++ and leveraging the unique unified main memory on ShenWei many-core architecture, we propose ShenWei Universal C/C++ (SWUC), a language extension to C/C++ that enables fluent programming acrossing the boundary of MPE and CPE. Through the use of several new attributes and compiler directives, users are able to write codes running on MPE and CPE in a single file. In case of C++, SWUC further support lambda expressions on CPEs, making it possible to have the code flow better matching the logical design. SWUC also manages to make the Athread library interfaces available, easing the learning curve for original ShenWei users. These powerful features together ensures SWUC to simplify the programming on ShenWei many-core processors and migration of existing C/C++ applications.

cs.PL↗

arXiv2020

CPM: A Large-scale Generative Chinese Pre-trained Language Model

Pre-trained Language Models (PLMs) have proven to be beneficial for various downstream NLP tasks. Recently, GPT-3, with 175 billion parameters and 570GB training data, drew a lot of attention due to the capacity of few-shot (even zero-shot) learning. However, applying GPT-3 to address Chinese NLP tasks is still challenging, as the training corpus of GPT-3 is primarily English, and the parameters are not publicly available. In this technical report, we release the Chinese Pre-trained Language Model (CPM) with generative pre-training on large-scale Chinese training data. To the best of our knowledge, CPM, with 2.6 billion parameters and 100GB Chinese training data, is the largest Chinese pre-trained language model, which could facilitate several downstream Chinese NLP tasks, such as conversation, essay generation, cloze test, and language understanding. Extensive experiments demonstrate that CPM achieves strong performance on many NLP tasks in the settings of few-shot (even zero-shot) learning. The code and parameters are available at https://github.com/TsinghuaAI/CPM-Generate.

cs.CL↗