dsh-linalg: DeepSeek Harness Linear Algebra Toolbox
dsh-linalg is a linear algebra toolbox for DeepSeek Harness, designed to address numerical errors that arise in large-model operations such as matrix multiplication and determinant computation. The plugin provides zero-dependency local exact arithmetic and supports capabilities such as matrix operations, solving linear systems, eigenvalue decomposition, and vector operations. After installation, users can configure the matrix dimension and the number of decimal places to retain via a configuration file. The plugin employs algorithms such as Gaussian elimination and the cyclic Jacobi method, using tolerance settings to ensure stable results.
Read MoreStudy Notes on Deep Learning I — Linear Algebra
This note covers various important concepts in machine learning, particularly those related to linear algebra. Below are some summaries and supplements to the content of the note: ### Fundamentals of Linear Algebra 1. **Matrices and Vectors**: Introduces matrices (arrays composed of multiple rows and columns) and vectors (essentially matrices with a single column or row). Emphasizes their importance in machine learning. 2. **Linear Combinations and Span**: - Linear Combination: Represented as $\sum_i x_i{\bf A}_{:,i}$. - Span (Note: The original content was cut off, so this is an assumption based on the context. If there was more specific content, please provide it for accurate translation.)
Read More