shivalikasingh commited on
Commit
2b6310b
·
verified ·
1 Parent(s): 4b9f841

Add 1 files

Browse files
Files changed (1) hide show
  1. README.md +62 -0
README.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ annotations_creators: []
3
+ language:
4
+ - en
5
+ language_creators: []
6
+ license: []
7
+ multilinguality:
8
+ - monolingual
9
+ pretty_name: 'adaption-databricks_dolly_15k'
10
+ size_categories:
11
+ - n<1K
12
+ source_datasets:
13
+ - 'extended|https://huggingface.co/datasets/databricks/databricks-dolly-15k'
14
+ tags:
15
+ - adaption
16
+ - preference-training
17
+ - animal-nature
18
+ - history
19
+ - sports
20
+ task_categories: []
21
+ task_ids: []
22
+ ---
23
+
24
+ ![banner](https://proteus-prod-public.s3.us-east-1.amazonaws.com/temp/f72607d5-590e-470d-b9c9-9508a6bfe14b.png)
25
+
26
+ This dataset is a remastered version of this [dataset](https://huggingface.co/datasets/databricks/databricks-dolly-15k) prepared using [Adaption's](https://adaptionlabs.ai/app/auth) Adaptive Data platform.
27
+
28
+ # adaption-databricks_dolly_15k
29
+
30
+ This dataset comprises over 15,000 open-source instruction-following records generated by Databricks employees across diverse categories like brainstorming, classification, and QA. Designed for training large language models, the English-language corpus features prompt-completion pairs that mimic human behavioral patterns outlined in the InstructGPT paper. It is licensed under Creative Commons Attribution-ShareAlike 3.0 for both academic and commercial use.
31
+
32
+ ### Dataset size
33
+
34
+ There are 200 data points in this dataset. This is a preference training dataset.
35
+
36
+ ### Quality of Remastered Dataset
37
+
38
+ The final quality is B, with a relative quality improvement of 58.0%.
39
+
40
+ ### Domain
41
+ - Animal-nature (16%)
42
+ - History (10%)
43
+ - Sports (10%)
44
+
45
+ ### Language
46
+ - English (100%)
47
+
48
+ ### Tone
49
+ - Informative (74%)
50
+ - Balanced (8%)
51
+ - Educational (6%)
52
+
53
+ ### Evaluation Results
54
+
55
+ - **Quality Gains:**
56
+ <img src="https://proteus-prod-public.s3.us-east-1.amazonaws.com/temp/fffdf4ad-90fa-4f33-acee-e6642cd9a127.png" alt="QualityGains" style="max-width: 50%; display: block; margin-left: auto; margin-right: auto;" />
57
+
58
+ - **Grade Improvement:**
59
+ <img src="https://proteus-prod-public.s3.us-east-1.amazonaws.com/temp/ebe98ac2-c07a-4c22-9442-e5841a2f1c7d.png" alt="Grade" style="max-width: 50%; display: block; margin-left: auto; margin-right: auto;" />
60
+
61
+ - **Percentile Chart:**
62
+ <img src="https://proteus-prod-public.s3.us-east-1.amazonaws.com/temp/c563e1a8-1c89-4242-aef9-84ac382ff738.png" alt="Percentile Chart" style="max-width: 50%; display: block; margin-left: auto; margin-right: auto;" />