metadata
language:
- fr
- en
size_categories:
- 1K<n<10K
task_categories:
- text-generation
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: eval
path: data/eval-*
- split: test
path: data/test-*
dataset_info:
features:
- name: category
dtype: string
- name: output
dtype: string
- name: query
dtype: string
- name: qid
dtype: int64
- name: fr_query
dtype: string
- name: fr_output
dtype: string
splits:
- name: train
num_bytes: 22352415.03147541
num_examples: 7866
- name: eval
num_bytes: 1810130.7367213115
num_examples: 637
- name: test
num_bytes: 1838547.2318032787
num_examples: 647
download_size: 16266132
dataset_size: 26001093
Dataset Card for "no_robots_enfr"
This is a filtered version of HuggingFaceH4/no_robots, then traduced to french with Deepl pro API, the best translation solution available on the market.
Our goal is to gather french data for one turn chatbot, on general subjects. We filtered few data from the original dataset:
- We kept only the one turn questions
- We took out any data where a system role is settle at the beginning, as our LLM will have a unique role that we don't have to define before a query.
- We kept the category information from the original dataset
Category | Number of Data | Mean Words (Query) | Mean Words (Output) |
---|---|---|---|
Brainstorm | 1120 | 35 | 217 |
Generation | 4560 | 35 | 177 |
Rewrite | 660 | 258 | 206 |
Open QA | 1240 | 12 | 73 |
Classify | 350 | 121 | 29 |
Summarize | 420 | 238 | 64 |
Coding | 350 | 55 | 124 |
Extract | 190 | 270 | 36 |
Closed QA | 260 | 217 | 22 |
---------------- | ---------------- | -------------------- | --------------------- |
General Dataset | 9150 | 71 | 150 |
Depending on our need we will filter those data by category to not inject hallicination in our fine-tuning.
The splits are made as each split have the same proportion of each categories: Train dataset size: 7866 Eval dataset size: 637 Test dataset size: 647