Datasets:
mike-conover-db
commited on
Commit
•
a9f8353
1
Parent(s):
c736d77
Initial commit.
Browse files
README.md
ADDED
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: mit
|
3 |
+
---
|
4 |
+
# Summary
|
5 |
+
`databricks-dolly-15k` is an open source dataset of instruction-following records generated by thousnads of Databricks empoyees in several
|
6 |
+
of the behavioral categories outlined in the InstructGPT paper, including brainstorming, classification, closed QA, generation, information
|
7 |
+
extraction, open QA, and summarization.
|
8 |
+
|
9 |
+
[LEGAL]
|
10 |
+
This dataset is licensed for commercial use ...
|
11 |
+
|
12 |
+
# Dataset Overview
|
13 |
+
Databricks employees were invited to create prompt / response pairs using Google Forms in each of eight different instruction categories, including the seven
|
14 |
+
outlined above as well as an open-ended free-form category. Employees were instructed to avoid using information from any source on the web with
|
15 |
+
the exception of Wikipedia, and explicitly instructed to avoid using generative AI in formulating instructions or responses. Examples of each behavior were
|
16 |
+
provided to motivate the types of questions and instructions appropriate to each category.
|
17 |
+
|
18 |
+
Halfway through the data generation process, participants were given the option of answering questions posed by other employees. They were
|
19 |
+
asked to rephrase the original question and only select questions they could be reasonably expected to answer correctly.
|
20 |
+
|
21 |
+
For questions in the XYZ categories employees were asked to provide a reference passage copied from Wikipedia. This text may contain bracketed
|
22 |
+
citations which we recommend users remove for downstream applications.
|
23 |
+
|
24 |
+
# Intended Uses
|
25 |
+
While immediately valuable for instruction fine tuning large language models, as a corpus of human-generated instruction prompts,
|
26 |
+
this dataset also presents a valuable opportunity for synthetic data generation in the methods outlined in the [Self-Instruct](https://arxiv.org/abs/2212.10560) paper.
|
27 |
+
For example, employee-generated prompts could be submitted as few-shot examples to a large open language model to generate a corpus of millions of examples of
|
28 |
+
instructions in each of the respective InstructGPT categories.
|
29 |
+
|
30 |
+
Likewise, both the instructions and responses present fertile ground for data augmentation. A paraphrasing model might be used to restate each prompt or short responses, with the
|
31 |
+
resulting text associated to the respective ground-truth sample. Such an approach might provide a form of regularization on the dataset that could allow for
|
32 |
+
more robust instruction-following behavior in models derived from these synthetic datasets.
|
33 |
+
|
34 |
+
# Dataset Limitations
|
35 |
+
[LEGAL]
|