DeepSeek-V2
DeepSeek-V2 is a state-of-the-art Mixture-of-Experts (MoE) language model introduced by DeepSeek-AI, characterized by its economical training and efficient inference capabilities. With a total of 236 billion parameters, of which only 21 billion are active per token, it supports a context length of up to 128K tokens. DeepSeek-V2 employs innovative architectures like Multi-head Latent Attention (MLA) for efficient inference by compressing the Key-Value (KV) cache and DeepSeekMoE for cost-effective training through sparse computation. This model significantly outperforms its predecessor, DeepSeek 67B, by saving 42.5% in training costs, reducing the KV cache by 93.3%, and enhancing generation throughput by 5.76 times. Pretrained on an 8.1 trillion token corpus, DeepSeek-V2 excels in language understanding, coding, and reasoning tasks, making it a top-tier performer among open-source models.
Learn more
BaseRT
BaseRT is a high-performance LLM inference runtime for Apple Silicon that lets developers pull models from Hugging Face, chat with them locally, or serve an OpenAI-compatible API from one CLI. Accelerated by hand-written Metal kernels, it is designed to deliver fast prefill and decode performance on M-series Macs, with published benchmarks showing up to 6.4× faster prefill than llama.cpp, 3.9× faster than MLX, and up to 1.33× faster decode. The basert CLI handles model downloading, conversion, interactive chat, serving, completion, benchmarking, inspection, and bundle signing. Its server supports chat, completions, embeddings, transcription, tool calls, continuous batching, paged KV caching, and prefix caching, while supported models can process text, vision, and audio. BaseRT uses its own .base model format with Q2–Q8 affine quantization, optional AWQ calibration, and signed bundles, and can convert GGUF, Hugging Face, and MLX checkpoints.
Learn more
Tensormesh
Tensormesh is a caching layer built specifically for large-language-model inference workloads that enables organizations to reuse intermediate computations, drastically reduce GPU usage, and accelerate time-to-first-token and latency. It works by capturing and reusing key-value cache states that are normally thrown away after each inference, thereby cutting redundant compute and delivering “up to 10x faster inference” while substantially lowering GPU load. It supports deployments in public cloud or on-premises, with full observability and enterprise-grade control, SDKs/APIs, and dashboards for integration into existing inference pipelines, and compatibility with inference engines such as vLLM out of the box. Tensormesh emphasizes performance at scale, including sub-millisecond repeated queries, while optimizing every layer of inference from caching through computation.
Learn more
PlatinumCache
DTS PlatinumCache C4 is a caching system developed by Data Transmission System Incorporation. DTS PlatinumCache C4 is a solution for the storage bottleneck problems. DTS PlatinumCache C4 is basically a caching system supporting Write-Back, Write-Through, Write-Only and Pre-fetching policies. The system uses RAMDISK as its cache. In android we used SD card as our target drive. DTS PlatinumCache C4 reads and writes data to/from SD card to increase throughput. It is designed in such a way to fulfill the needs of every category of clients ranging from small to large enterprises. Performance becomes greatly enhanced because data transfer occurs from Cache which is RAM. MCell-II SSD first in the industry. SSD version of the hybrid memory disk. The via DRAM, 30,000 IOPS Random Read, 26,000 IOPS Ultra-fast access Random Write. The DTS chip, manage the number of writes. Achieve a long life. (DTS PlatinumCache that the chunk size to write to the SSD.
Learn more
