Model Training

Use your deployed MSP management API for management operations and its inference Gateway for model requests. Check the deployed version and required permissions. Example addresses are placeholders.

Prepare datasets, fine-tune models, and evaluate performance on your own clusters.

The Model Training module covers the full training pipeline across three areas:

Datasets: Create datasets by manually uploading files in JSONL or TAR format, or use AI Dataset Preparation to automatically label and structure raw data. Datasets can be shared across training and evaluation tasks.

Model Training: Fine-tune base models using three methods. SFT (Supervised Fine-Tuning) adapts models with labeled examples for specific tasks. DPO (Direct Preference Optimization) aligns models with human preferences using preference pairs. CPT (Continual Pre-Training) injects domain knowledge by continuing pre-training on domain-specific corpora.

Evaluations: Measure model performance using Benchmark Evaluation against standardized public datasets (e.g., MMLU, GSM8K, HumanEval), or AI Auto Evaluation with custom datasets and LLM-as-Judge scoring.

Create a Dataset

Upload or AI-generate datasets for training and evaluation.

Learn More

Fine-Tune a Model

Train a base model to your domain using SFT, DPO, or CPT.

Learn More

Evaluate a Model

Evaluate models' performance with scientific methods.

Learn More