Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.benchgen.com/llms.txt

Use this file to discover all available pages before exploring further.

Fine-tune a Model

Fine-tuning adapts a base model to your specific task using a LoRA (Low-Rank Adaptation) adapter. Train manages the compute — you just configure the run.

Prerequisites

  • A dataset uploaded or available in Train (see Upload a dataset)
  • A base model selected (see supported base models below)

Steps

  1. Go to Train → Fine-tune → New Run.
  2. Select a base model from the supported model list.
  3. Select a dataset from your uploaded datasets.
  4. Configure LoRA parameters:
ParameterDescriptionRecommended starting value
rankLoRA rank — controls adapter capacity16
alphaScaling factor for the adapter32
target_modulesWhich layers to adaptq_proj, v_proj
dropoutRegularization0.05
  1. Configure training hyperparameters:
ParameterDescriptionRecommended starting value
learning_rateStep size for gradient updates2e-4
epochsNumber of full passes over the dataset3
batch_sizeExamples per gradient step4
max_seq_lengthMaximum token length per example2048
  1. Click Start run.
  2. Monitor the run — loss curve and sample outputs update in real time.
  3. When complete, the adapter is saved and ready to merge.

Next Steps