Skip to content

Pathfinder is a Fully Self-Supervised Language model tuning technique.

Notifications You must be signed in to change notification settings

AI4Chem/UnsupervisedFinetune

Repository files navigation

PathFinder: Towards Self-Supervised Evolution LLM without Human Feedbacks

Pathfinder is a Fully Self-Supervised Language model tuning technique.

In the quest to leverage the vast amounts of unlabeled data available, we introduce PathFinder, a novel, fully self-supervised fine-tuning algorithm for Large Language Models (LLMs). PathFinder necessitates only the LLM's inherent ability to follow basic instructions, enabling it to exploit any unlabeled dataset of questions, which can also be autonomously sampled by the model itself. This method advances the model's comprehension and generative capabilities without relying on labeled data, thus paving the way for fully autonomous and efficient LLM evolution.

Usage

Install conda dependencies in conda.yaml configure Accelerate like accelerate.yaml

cd LLaMa-Factory
accelerate launch src/train_bash.py     --stage sft     --do_train     --model_name_or_path google/gemma-2b-it     --dataset mathinstruct    --template gemma     --finetuning_type lora --lora_target all     --output_dir path_to_math_checkpoint      --per_device_train_batch_size 1     --gradient_accumulation_steps 2     --lr_scheduler_type cosine     --logging_steps 1     --save_steps 50     --learning_rate 5e-5     --num_train_epochs 1     --plot_loss     --bf16  --upcast_layernorm=true --ddp_find_unused_parameters=false --flash_attn --neftune_noise_alpha 5 --preprocessing_num_workers 8 --overwrite_output_dir

Credit

Zhang Di @ FDU & ShanghaiAILab

About

Pathfinder is a Fully Self-Supervised Language model tuning technique.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published