Search Results for "lepton-optimizer"

...AMD MIVisionX delivers highly optimized open-source implementation of the Khronos OpenVX™ and OpenVX™ Extensions along with Convolution Neural Net Model Compiler & Optimizer supporting ONNX, and Khronos NNEF™ exchange formats. The toolkit allows for rapid prototyping and deployment of optimized computer vision and machine learning inference workloads on a wide range of computer hardware, including small embedded x86 CPUs, APUs, discrete GPUs, and heterogeneous servers. AMD OpenVX is a highly optimized open-source implementation of the Khronos OpenVX™ 1.3 computer vision specification. ...

Polars is a high-performance, multi-language DataFrame library built in Rust using Apache Arrow. It delivers blazing-fast, vectorized, and parallel data manipulation with both eager and lazy execution, making it an excellent tool for data processing in Python, Rust, Node.js, R, and SQL contexts.

...Binaryen's internal IR uses compact data structures and is designed for completely parallel codegen and optimization, using all available CPU cores. Binaryen's IR also compiles down to WebAssembly extremely easily and quickly because it is essentially a subset of WebAssembly. Binaryen's optimizer has many passes (see an overview later down) that can improve code size and speed. These optimizations aim to make Binaryen powerful enough to be used as a compiler backend by itself.

FairScale is a collection of PyTorch performance and scaling primitives that pioneered many of the ideas now used for large-model training. It introduced Fully Sharded Data Parallel (FSDP) style techniques that shard model parameters, gradients, and optimizer states across ranks to fit bigger models into the same memory budget. The library also provides pipeline parallelism, activation checkpointing, mixed precision, optimizer state sharding (OSS), and auto-wrapping policies that reduce boilerplate in complex distributed setups. Its components are modular, so teams can adopt just the sharding optimizer or the pipeline engine without rewriting their training loop. ...

...We provide reference implementations of various sequence modeling papers. Recent work by Microsoft and Google has shown that data parallel training can be made significantly more efficient by sharding the model parameters and optimizer state across data parallel workers. These ideas are encapsulated in the new FullyShardedDataParallel (FSDP) wrapper provided by fairscale. Fairseq can be extended through user-supplied plug-ins. Models define the neural network architecture and encapsulate all of the learnable parameters. Criterions compute the loss function given the model outputs and targets. ...

Zopfli is a compression library and command-line tool that produces exceptionally small DEFLATE, zlib, and gzip streams by spending more CPU time to search for better encodings. It keeps strict compatibility with the ubiquitous DEFLATE format, so outputs can be decompressed by any standard tool or browser. The encoder performs exhaustive block splitting and greedy but thorough match searching to shave extra bytes off assets, which is ideal for web content and firmware where size matters more...