Search Results for "mla"

DeepSeek-V3 is a robust Mixture-of-Experts (MoE) language model developed by DeepSeek, featuring a total of 671 billion parameters, with 37 billion activated per token. It employs Multi-head Latent Attention (MLA) and the DeepSeekMoE architecture to enhance computational efficiency. The model introduces an auxiliary-loss-free load balancing strategy and a multi-token prediction training objective to boost performance. Trained on 14.8 trillion diverse, high-quality tokens, DeepSeek-V3 underwent supervised fine-tuning and reinforcement learning to fully realize its capabilities. ...

OpenMythos is an experimental, open-source implementation that attempts to reconstruct a hypothesized architecture behind advanced language models using a design called a Recurrent-Depth Transformer. The project explores the idea that instead of stacking hundreds of unique transformer layers, a smaller set of layers can be reused iteratively during inference to achieve deeper reasoning without increasing parameter count. It divides computation into three main stages, including a...

infiAgent is an open-source AI agent framework for building powerful, long-running autonomous agents capable of tackling complex tasks without collapsing under growing context or tool invocation histories. Designed as a “Multi-Level Agent” (MLA) system, it externalizes persistent state to the file system so that agents can operate over unlimited runtime without the need for token-intensive context compression, enabling workflows such as research paper drafting, experiments, coding, and document generation to run reliably. The framework uses a serial multi-agent hierarchy where specialized agents coordinate in tree-structured paths for clear task delegation and minimal tool conflicts, while batch file operations and persistent workspaces ensure reproducibility and traceability. ...