How-to Guides#
Train an LLM using Maximal Update Parameterization
Learn how to enable μP when training models on the Cerebras Wafer-Scale cluster
Extend Context Length Using Position Interpolation
Learn how to use Position Interpolation to enable models using RoPE or ALiBi to efficiently extend their context length.
Train an LLM with a large or small context window
Learn how to use the CS-X to train an LLM with a large or small context window
Instruction fine-tune an LLM
Learn how to fine-tune LLMs on datasets with instructions and corresponding responses
Train a model with weight sparsity
Learn how to train a model with weight sparsity to achieve a sparse model that requires fewer FLOPs to train and fewer parameters to store
Restart a dataloader
Learn how to resume training from the same point in the input-generating dataloader
Port a trained and fine-tuned model to Hugging Face
Learn how to port an LLM model trained in the Cerebras’s Wafer-Scale Cluster to Hugging Face to generate outputs
Port a Hugging Face model to Cerebras Model Zoo
Learn how to port a Hugging Face model to the Cerebras Model Zoo to generate outputs
Control numerical precision level
Learn how to control the level of numerical precision used for training runs for large NLP models
Enable Dynamic Loss Scaling
Learn how to enable dynamic loss scaling using Cerebras's custom PyTorch module.
Run Cerebras Model Zoo on a GPU
Learn how to run models in the Cerebras Model Zoo on GPUs and which packages to install