Skip to content

Latest commit

 

History

History
58 lines (52 loc) · 1.58 KB

README.md

File metadata and controls

58 lines (52 loc) · 1.58 KB

Inference

If you want to reproduce our model's results on Flores-101 from our main experiment table, you can refer to the following steps.

Environment

We used Python 3.9 to run our inference code on A100. Please follow the steps below to install the Python environment.

pip install -r requirements.txt

Data

cd ../../data/
wget https://dl.fbaipublicfiles.com/flores101/dataset/flores101_dataset.tar.gz
tar -xvzf flores101_dataset.tar.gz

Configure

You can adjust the following parameters in the config file(./configs/flores101.yaml) according to your needs.

model_path: 
   base_model: ../../model/LLaMAX2-7B-Alpaca/  ## your model path
   lora: null
   torch_dtype: bf16  ## torch dtype
   llama_type: llama2  ## llama2 or llama3 model
dataset:
   loader: flores101 ## dataset name
   path: ../../data//flores101_dataset/devtest/ ## dataset root
   input_file: eng.devtest ## data path
   inst_file: ./prepare/instruct_inf.txt  ## instruction file
   lang_instruction: ./prepare/language_name_code_pair.xlsx  ## a table of language IDs and their full names
   input_extra_file: null
   lang_pair: en-zh  ## translation pair
   inst_placeholder: null
   inst_fix: true
   inst_index: 0
   labels: null
   inst_with_input: true 
   input_size: -1
   use_match: false

generat:
    beam_size: 4   ## beam size
    temperature: 0.1
    batch_size: 8  ## batch size
    template: prompt_input
    search: beam 
    do_sample: false
    
output:
    subpath: ""
    output_file_prefix: generation_results

Usage

cd inference
python eval.py --cfg ./configs/flores101.yaml