Fu01978 commited on
Commit
979e026
·
verified ·
1 Parent(s): da39ae9

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +81 -0
README.md ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ task_categories:
4
+ - text-generation
5
+ size_categories:
6
+ - 10K<n<100K
7
+ language:
8
+ - en
9
+ tags:
10
+ - chat
11
+ - conversation
12
+ - instruction-tuning
13
+ - fanfiction
14
+ - ao3
15
+ pretty_name: AO3 Chat
16
+ ---
17
+
18
+ # ao3_chat
19
+
20
+ ## Overview
21
+
22
+ A reformatted version of the dataset
23
+ [**`midwestern-simulation-active/ao3_random_subset`**](https://huggingface.co/datasets/midwestern-simulation-active/ao3_random_subset), converted into
24
+ chat-style and instruction-tuning examples suitable for training
25
+ chat-oriented language models.
26
+
27
+ Each example consists of:
28
+ - a **system** prompt derived from AO3 metadata
29
+ - a **user** prompt requesting the contents of a specific work
30
+ - an **assistant** response containing the full text (or a chunk thereof)
31
+
32
+ This dataset is intended for supervised fine-tuning (SFT) of chat models.
33
+
34
+ ---
35
+
36
+ ## Structure
37
+
38
+ The repository contains two representations of the same data:
39
+
40
+ ### `jsonl/`
41
+ Line-delimited JSON files
42
+
43
+ ### `parquet/`
44
+ Columnar Parquet files containing the same fields, suitable for
45
+ efficient loading with PyArrow, Pandas, or Hugging Face Datasets.
46
+
47
+ ## Prompt Format
48
+
49
+ ### System
50
+ Metadata extracted from AO3 entries (rating, fandoms, relationships,
51
+ characters, tags, etc.).
52
+
53
+ ### User Prompt
54
+ A request to provide the full text of a specific fanfiction work by title.
55
+
56
+ ### Assistant Response
57
+ The original fanfiction text, optionally split into multiple parts for
58
+ long works.
59
+
60
+ ## Source
61
+
62
+ Original data:
63
+
64
+ [`midwestern-simulation-active/ao3_random_subset`](https://huggingface.co/datasets/midwestern-simulation-active/ao3_random_subset)
65
+
66
+ This dataset preserves the original text content and metadata, while
67
+ changing only the formatting.
68
+
69
+ ## Notes
70
+
71
+ * This dataset contains copyrighted material.
72
+
73
+ * It is provided for research and experimentation purposes.
74
+
75
+ * Users are responsible for complying with AO3’s terms of service and
76
+ applicable copyright laws.
77
+
78
+ ## Intended Use
79
+
80
+ - Chat model fine-tuning
81
+ - Long-form text generation research