metadata
dataset_info:
features:
- name: type
dtype: string
- name: messages
dtype: string
- name: answer
dtype: string
- name: answer_english
dtype: string
- name: lang
dtype: string
- name: hi_answer
dtype: string
- name: mt_question
sequence: string
splits:
- name: train
num_bytes: 756440
num_examples: 734
download_size: 219666
dataset_size: 756440
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Evaluation on hindi and english prompts borrowed from teknimum, airoboros, https://huggingface.co/datasets/HuggingFaceH4/mt_bench_prompts, https://huggingface.co/datasets/ai4bharat/human-eval and other sources
Mainly used to evalaution on written tasks through LLM JUDGE
https://github.com/lm-sys/FastChat/blob/main/fastchat/llm_judge/README.md