metadata
license: apache-2.0
datasets:
- OpenAssistant/oasst_top1_2023-08-25
language:
- en
pipeline_tag: text-generation
base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-955k-token-2T
TinyLlama/TinyLlama-1.1B-intermediate-step-955k-token-2T finetuned using OpenAssistant/oasst_top1_2023-08-25 dataset.
Trained for 5 epochs using Qlora. Adapter is merged.
SFT code: https://github.com/habanoz/qlora.git
Command used:
accelerate launch $BASE_DIR/qlora/train.py \
--model_name_or_path $BASE_MODEL \
--working_dir $BASE_DIR/$OUTPUT_NAME-checkpoints \
--output_dir $BASE_DIR/$OUTPUT_NAME-peft \
--merged_output_dir $BASE_DIR/$OUTPUT_NAME \
--final_output_dir $BASE_DIR/$OUTPUT_NAME-final \
--num_train_epochs 5 \
--logging_steps 1 \
--save_strategy steps \
--save_steps 75 \
--save_total_limit 2 \
--data_seed 11422 \
--evaluation_strategy steps \
--per_device_eval_batch_size 4 \
--eval_dataset_size 0.01 \
--eval_steps 75 \
--max_new_tokens 1024 \
--dataloader_num_workers 3 \
--logging_strategy steps \
--do_train \
--do_eval \
--lora_r 64 \
--lora_alpha 16 \
--lora_modules all \
--bits 4 \
--double_quant \
--quant_type nf4 \
--lr_scheduler_type constant \
--dataset oasst1-top1 \
--dataset_format oasst1 \
--model_max_len 1024 \
--per_device_train_batch_size 4 \
--gradient_accumulation_steps 4 \
--learning_rate 1e-5 \
--adam_beta2 0.999 \
--max_grad_norm 0.3 \
--lora_dropout 0.0 \
--weight_decay 0.0 \
--seed 11422 \
--gradient_checkpointing \
--use_flash_attention_2 \
--ddp_find_unused_parameters False