Liquid AI released an experimental 279.5-million-parameter draft model, DSpark, for its LFM2.5-VL-3B vision-language model on September 24th, which the company says cuts decoding time by up to 3.13x o…
OpenBMB released MiniCPM5-2B, a 2.52 billion parameter open-weight language model that posts an average benchmark score of 53.9 across nine capability axes, ahead of Qwen3.5-4B (51.1), granite-4.2-3B …
DeepSeek's DSpark speculative decoding technique, which combines parallel drafting with a lightweight sequential component, can improve local LLM generation speed on the same GPU, with DeepSeek report…
NVIDIA's DGX Spark, powered by the GB10 superchip with 128 GB of unified LPDDR5X memory, achieves only 11.5 tokens per second running Qwen3.8-27B at Q4_K_XL quantization, limited by 273 GB/s memory ba…
Liquid AI released DSpark draft model checkpoints for its LFM2.5 family, claiming up to 3.18x throughput improvement on GPU and up to 2.87x on-device, with day-one support for llama.cpp and SGLang. Th…
A developer detailed the process of running Qwen3.8-27B on dual RTX 3090s without NVLink under WSL2, achieving 170-210 tok/s on code/JSON. The setup required specific CUDA 13.0 toolchain, SGLang 0.5.1…
DeepSeek AI released DeepSeek-V4-Pro-0813, the official non-preview version of its open-weight coding and agentic model, reporting gains over the preview on every listed benchmark. On Terminal Bench 2…
DeepSeek released DeepSeek-V4-Pro-0813, the production successor to its V4 Pro preview, under an MIT license with open weights, adding a DSpark speculative decoding module and stronger agentic benchma…
DeepSeekV4SSD, an experimental app from developer yanun0323, streams routed experts from SSD to run all 284 billion parameters of DeepSeek-V4-Flash-0731 on an M-series Mac with about 30 GB of memory, …
DeepSeek released DeepSeek-V4-Flash-0731 on Hugging Face and moved its V4-Flash API to public beta on July 31, 2026, claiming major gains in agentic and coding benchmarks from re-post-training rather …
VLLM announces efficient day-0 support for Moonshot AI's Kimi K3, a 2.8-trillion-parameter Mixture-of-Experts model, achieving up to 370 tokens per second with speculative decoding on 16 NVIDIA GB300 …
DSpark introduces a sharding strategy that reduces GPU communication overhead to solve the KV cache memory bloat in long-context LLM inference, targeting latency spikes in distributed setups. The appr…
Researchers propose DeLS-Spec, a speculative decoding method that decouples long- and short-context experts to accelerate LLM inference. By adding a lightweight local head to the existing DFlash model…
DeepSeek launched DSpark, an open-source speculative decoding module that boosts AI inference speed by up to 400% on existing chips, potentially reducing demand for Nvidia's high-end accelerators. The…
Deepseek's new DSpark framework boosts per-user AI response speed by 60 to 85 percent by using a small model to propose token candidates that a larger model checks in batches, reducing dependence on U…
Chinese AI firm DeepSeek open-sourced DSpark, a speculative decoding system that accelerates large language model inference by up to 85% without altering output quality, releasing it under the MIT lic…
DeepSeek released DSpark, an open-source inference framework that accelerates AI model generation by up to 85% without hardware upgrades, challenging U.S. export controls on advanced chips. The framew…
Chinese AI startup DeepSeek unveiled DSpark, a speculative decoding framework that speeds up AI inference by up to 85% while reducing costs and chip strain. The module uses a lightweight draft model a…
An engineer analyzed speculative decoding for LLM inference acceleration, finding that while it offers lossless quality by mathematically matching autoregressive output, its speedup depends on accepta…
DeepSeek's DSpark paper reframes speculative decoding by grafting a speculative head directly onto the target model rather than training a separate draft model. The technique reduces layer duplication…