DeepSeek-V4-Flash-0731: surpasses Fable-5, Sol & Kimi-K3 on Chess Benchmark

TensorSharp now supports DSpark on Deepseek v4 Flash 0731, demonstrating significant performance improvements. Benchmarks on 4x Nvidia A40 GPUs with CUDA 12.8 show DSpark enhancing generation speeds across various tasks. For instance, short generation (53 tokens) saw a 1.74x speedup, long generation (512 tokens) a 1.53x speedup, and 10K-token document processing a 2.03x speedup. TensorSharp is an open-source inference engine for GGUF LLMs, featuring CUDA, Vulkan, Metal, and OpenAI-compatible APIs.