Chinese-LLaMA-Alpaca-2 of LLMs: source code interpretation (run_clm_pt_with_peft.py file) - model training pre-work (parameter analysis + configuration log) → model initialization (detecting whether there is a trained checkpoint + loading pre-trained model and tokenizer) → Data preprocessing (processing [tokenization + chunking] + segmentation txt data set) → Optimize model configuration ( quantization module + matching model vocabulary size and tokenizer + initialization PEFT model [LoRA] + gradient accumulation checkpointing, etc.) → model training (continue Training + evaluation indicators + automatic saving of intermediate training results)/model evaluation (+PPL indicators)
Table of contents