Orchestra-Research

Framework for state-of-the-art sentence, text, and image embeddings.

Language-independent tokenizer treating text as raw Unicode.

Fast structured generation and serving for LLMs with RadixAttention prefix caching.

Simple Preference Optimization for LLM alignment.

Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed…

Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework.

Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques.

State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers.

Provides guidance for experiment tracking with SwanLab.

Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with…

Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency.

Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms.