zyds/transformers-code

Covers comprehensive Transformers training workflows across foundational components (Pipeline, Tokenizer, Model, Datasets, Trainer) and nine NLP task implementations (NER, machine reading comprehension, text similarity, dialogue systems, summarization). Integrates PEFT library for parameter-efficient fine-tuning methods (LoRA, QLoRA, Prompt-tuning, P-tuning), bitsandbytes for 4/8-bit quantization training, and Accelerate for distributed training with DeepSpeed support. Targets PyTorch 2.2+ with compatible versions of transformers 4.42+, datasets, and evaluation frameworks.

3,853 stars. No commits in the last 6 months.