Awesome GitHub RepositoriesDataset Blending

gpt-neox is a distributed training system and framework for building large-scale autoregressive language models. It implements the transformer architecture and provides a toolkit for training models with billions of parameters by distributing weights across compute clusters. The framework distinguishes itself through extensive support for distributed model parallelism, including pipeline and sequence parallelism, to overcome single-device memory limits. It further supports sparse model architectures using a mixture of experts system with Sinkhorn-based routing. The project covers a broad ran

Torchtune is a PyTorch-native library for fine-tuning, aligning, and quantizing large language models. It provides a config-driven system for instantiating components, orchestrating distributed training, and managing parameter-efficient fine-tuning with quantization support, all through YAML-based configurations and command-line overrides. The library distinguishes itself through its comprehensive post-training workflow orchestration, combining supervised fine-tuning, preference optimization (DPO, PPO, GRPO), knowledge distillation, and quantization-aware training in a single configurable pip

Dieses Projekt ist ein Alignment-Framework und eine Pipeline-Suite für das Training von Sprachmodellen mittels Supervised Fine-Tuning und Preference Optimization. Es bietet Tools für die Ausführung von großskaligem, verteiltem Training über mehrere GPUs und Compute-Nodes hinweg, ergänzt durch ein System zur Messung von Hilfsbereitschaft und Dialogqualität durch Single-Turn- und Multi-Turn-Benchmarks. Das Framework enthält spezialisierte Tools für Direct Preference Optimization, um das Modellverhalten mithilfe gepaarter Daten ohne separates Reward-Modell zu verfeinern. Es unterstützt zudem Constitutional AI Alignment sowie das Training von Reward-Modellen zur Bewertung und Einstufung von Antworten basierend auf Präferenzkriterien. Das Projekt deckt umfassendere Funktionen für das Blending und Mischen von Datensätzen, Parameter-effizientes Fine-Tuning via Low-Rank Adaptation sowie Rejection Sampling Optimization ab. Es verwaltet den Trainings-Lebenszyklus durch konfigurationsgesteuerte Rezepte und bietet Systeme für das Streaming von Echtzeit-Leistungsmetriken an externe Dashboards.

OpenVLA is a vision-language-action model and framework designed for general-purpose robotic manipulation. It provides a robotic policy training framework and a control inference engine that map visual and textual inputs to robotic control actions, enabling zero-shot instruction following on hardware. The project includes a robotics dataset pipeline for standardizing diverse trajectory data and managing dataset mixtures. It supports large-scale model training through distributed GPU compute and sharded data parallelism, alongside parameter-efficient adaptation for fine-tuning models to new ta