QLoRA Fine-Tuning
CLI Equivalent: apr finetune --method qlora --rank 8 --alpha 16 model.apr
What This Demonstrates
Quantized LoRA (QLoRA) fine-tuning that keeps base model weights in 4-bit precision while training full-precision LoRA adapters. Enables fine-tuning of large models on limited VRAM.
Run
cargo run --example finetune_qlora
Key APIs
LoRAConfig::new(rank, alpha).target_qv_projections()-- configure adapter dimensionsQuantization::Int4-- quantize base weights to 4-bitLoRALayer::new(quantized_base, d_out, d_in, rank, alpha)-- attach adapters to quantized base.trainable_params()-- only adapter weights are trainableMemoryPlanner::estimate_vram(config)-- predict peak VRAM usage