Customize

• LoRA and X-LoRA adapters: attach fine-tuned adapters to a base model.
• AnyMoE: mix multiple experts at inference time.
• MatFormer: elastic model sizing from a MatFormer-trained checkpoint.
• Chat templates: when auto-detection is wrong or missing.
• Sampling parameters: temperature, top-k, top-p, min-p, DRY, and their interactions.