metadata
license: apache-2.0
datasets:
- pankajmathur/WizardLM_Orca
language:
- en
pipeline_tag: text-generation
Mistral 7b Reverse Instruct
This LoRA Adapter is fine tuned to reverse engineer the original prompt of a given LLM output/response.
- base_model: mistralai/Mistral-7B-v0.1 (=checkpoint-v1)
- base_model: mistralai/Mistral-7B-v0.2 (>=checkpoint-v2)
For convinience the latest model export is provided under /latest_model_export
Response Format
"[INST]\n### System:\n{system}\n### Instruction:\n{instruction}\n[/INST]\n"
Prompt Template
"\n### System:\nYou craft instructions for generating the given output through reverse engineering.\n### Instruction:\nDecipher the steps used to produce the given output and articulate a refined set of instructions (System & Instruction).\n### OUTPUT:\n {output}"
(use the template without the " ")
Training Dataset
About 21k items of the following datasets were used. (mostly coding-like tasks were removed)
wget https://raw.githubusercontent.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM/main/data/alpaca_gpt4_data.json
wget https://raw.githubusercontent.com/teknium1/GPTeacher/main/Roleplay%20Supplemental/roleplay-instruct-v2.1.json
wget https://huggingface.co/datasets/pankajmathur/WizardLM_Orca/resolve/main/wizardlm_orca.json
Training Procedure
CUDA_VISIBLE_DEVICES=0 WANDB_DISABLED=True python LLaMA-Factory/src/train_bash.py \
--stage sft \
--model_name_or_path model_name_or_path \
--checkpoint_dir checkpoint_dir \
--flash_attn \
--shift_attn \
--neftune_noise_alpha 5 \
--do_train \
--dataset default \
--template vanilla \
--finetuning_type lora \
--lora_target q_proj,v_proj \
--output_dir path_to_sft_checkpoint \
--overwrite_cache \
--per_device_train_batch_size 1 \
--gradient_accumulation_steps 1 \
--lr_scheduler_type cosine \
--logging_steps 10 \
--save_steps 100 \
--learning_rate 5e-5 \
--num_train_epochs 3.0 \
--plot_loss \
--fp16 \
--overwrite_output_dir \
--cutoff_len 2048 \
--quantization_bit 4
Training Time
- v1: ~12h on Kaggle's P100 GPU
- v2: >30h on Kaggle's T4 x2
Framework versions
- LLaMA-Factory