Blog

SWE-bench +

❖ SWE-bench is a benchmark for evaluating large language models on real world software issues collected from GitHub. Given a codebase and an issue, a language model is tasked with generating a patch that resolves the described problem. SWE-bench Verified is a human-validated subset that more reliably evaluates AI models’ ability to solve issues. International Olympiad in Informatics (IOI) competition features standardized and automated grading.

❖ DeepSeek Sparse Attention (DSA) was a breakthrough for long-context models, reducing core attention complexity from O(L²) to O(Lk), but its lightning indexer remained an expensive O(L²) tax at every layer. IndexCache eliminates up to 75% of indexer computations in DSA through cross-layer index reuse — achieving up to 1.82× prefill speedup and 1.48× decode speedup with negligible quality degradation.

TurboQuant

❖ TurboQuant compresses vectors by rotating them into a predictable distribution, then snapping to a precomputed grid. Step through the animation below to see every stage of TurboQuant: normalize, rotate, quantize, reconstruct. Hover over any point for exact coordinates. On step 0, click to add or remove points.

❖ This is the introduction of Attention Residuals (AttnRes), a drop-in replacement for standard residual connections in Transformers that enables each layer to selectively aggregate earlier representations via learned, input-dependent attention over depth.

Optimizing DSA in TensorRT LLM

❖ The open-sourced DeepSeek-V3.2 series models proposed a new architecture with a fine-grained sparse attention mechanism, called DeepSeek Sparse Attention (DSA). It can help the DeepSeek-V3.2 model achieve better efficiency, especially in long sequence scenarios. Although DSA uses a lightweight indexer for prediction, realizing actual speedup from attention sparsity is still challenging.

CUDA Graphs in vLLM

❖ This write-up introduces the new CUDA Graphs modes in vLLM v1. To summarize: Added flexible cudagraph_mode configuration. Made full CUDA Graphs support orthogonal to compilation. Introduced a CUDA Graphs dispatcher as a central controller that picks the desired runtime mode and CUDA Graphs per batch automatically.

❖ The GLM-4 series models are foundation models designed for intelligent agents. GLM-4.7 has 355 billion total parameters with 32 billion active parameters, while GLM-4.5-Air adopts a more compact design with 106 billion total parameters and 12 billion active parameters. GLM-4.x models unify reasoning, coding, and intelligent agent capabilities to meet the complex demands of intelligent agent applications.

Qwen3-VL

❖ Qwen3-VL is the most powerful vision-language model in the Qwen series to date. It delivers comprehensive upgrades across the board: superior text understanding & generation, deeper visual perception & reasoning, extended context length, enhanced spatial and video dynamics comprehension, and stronger agent interaction capabilities. Available in Dense and MoE architectures that scale from edge to cloud, with Instruct and reasoning‑enhanced Thinking editions for flexible, on‑demand deployment.