OLMo Core
An open-source PyTorch library for OLMo model training, with official training scripts and inference via Hugging Face Transformers or vLLM.
OLMo Core is a Python library for researchers and developers building and training OLMo language models. Its PyTorch components and official training scripts suit teams working with model training code on their own hardware or servers.
- Official model training: Training scripts and model cards cover OLMo-2 32B and OLMo-3 7B and 32B models.
- Training customization: Adjust training settings and continue annealing from an existing checkpoint with a dedicated script.
- Inference runtimes: Run OLMo checkpoints through Hugging Face Transformers, or use vLLM for high-throughput inference and offline batch processing.
OLMo Core is open source under Apache 2.0 and depends on PyTorch for its operating system and hardware support. Training uses PyTorch's torchrun launcher, with Beaker as an optional alternative for users who have access to it. Separate OLMo Eval and olmes projects provide evaluation tools for OLMo models.
Logotyper
