Search Results for "c programming framework" - Page 2

...We developed efficient, model-parallel (tensor, sequence, and pipeline), and multi-node pre-training of transformer based models such as GPT, BERT, and T5 using mixed precision. Megatron is also used in NeMo Megatron, a framework to help enterprises overcome the challenges of building and training sophisticated natural language processing models with billions and trillions of parameters. Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.

RTP-LLM is an open-source large language model inference acceleration engine developed by Alibaba to provide high-performance serving infrastructure for modern LLM deployments. The system focuses on improving throughput, latency, and resource utilization when running large models in production environments. It achieves this by implementing optimized GPU kernels, batching strategies, and memory management techniques tailored for transformer inference workloads. The framework is designed for...

ToRA is an open-source framework developed by Microsoft for building tool-integrated reasoning agents powered by large language models. The project focuses on improving the ability of AI systems to solve complex mathematical and analytical problems by combining natural language reasoning with external computational tools. Instead of relying solely on text generation, the system dynamically invokes tools such as symbolic solvers or programming libraries when deeper computation is required. ...

llama.go is like llama.cpp in pure Golang. The code of the project is based on the legendary ggml.cpp framework of Georgi Gerganov written in C++ with the same attitude to performance and elegance. Both models store FP32 weights, so you'll needs at least 32Gb of RAM (not VRAM or GPU RAM) for LLaMA-7B. Double to 64Gb for LLaMA-13B.