caddyserver/caddy
Stars: 49.1k License: Apache-2.0
Caddy is a scalable server platform that uses TLS by default.
- Simple and easy to use configuration method
- Supports powerful local JSON configuration and dynamic JSON API configuration
- Automatic HTTPS functionality is provided by default and supports ZeroSSL and Let's Encrypt certificate authorities.
- Can work with other Caddy instances to run in a cluster
- It is highly scalable and can implement various functions through plug-ins without being bloated.
- No external dependencies required (not even libc), runs anywhere
- Written in Go language, it has higher memory safety performance than other servers.
ddbourgin/numpy-ml
Stars: 12.2k License: GPL-3.0
numpy-ml is a collection of machine learning algorithms implemented in NumPy. Although it is not efficient, it is relatively easy to read. The main features of this project include providing various models and tool functions to support machine learning tasks.
- Provides multiple common machine learning models, such as Gaussian mixture models, hidden Markov models, latent Dirichlet allocation, etc.
- Supports neural network related operations, such as fully connected layers, LSTM, convolution, etc.
- Including tree models (decision tree, random forest), linear regression and classification (ridge regression, logistic regression) and non-parametric methods (kernel regression, nearest neighbor classification and regression), etc.
- Provides data preprocessing functions, such as signal transformation (Fourier transform and discrete cosine transform), feature encoding/decoding, and text processing.
- Other auxiliary tool functions include similarity calculation kernel functions and distance measurement methods.
cameron314/concurrentqueue
Stars: 8.1k License: NOASSERTION
moodycamel::ConcurrentQueue is an industrial-grade lock-free queue for C++.
- Extremely fast performance
- Single header file implementation for easy integration into projects
- Fully thread-safe lock-free queue that can be used concurrently from any number of threads at the same time
- C++11 implementation, moving elements when possible (instead of copying)
- Templated design avoids the trouble caused by only dealing with pointer type objects - memory management is completed by the library itself
- No artificial restrictions on element types or maximum number
- Memory space can be pre-allocated at one time or dynamically allocated as needed.
- Full platform support (implemented through standard C++11 primitives)
haotian-liu/LLaVA
Stars: 4.7k License: Apache-2.0
LLaVA is a large-scale language and visual assistant project whose main function is to adjust visual instructions to implement large-scale language and visual models with GPT-4 level capabilities.
- Supports LoRA training, 4-/8-bit inference, higher resolution, etc.
- LLaVA Bench for benchmarking open visual chat is provided, with results from Bard and Bing-Chat
- Supports the use of RTX 3090 and RTX A6000 for training verification
- LLaVA-13B that can run on GPU with only 12GB VRAM (quantified)