Datasets:

Modalities:
Text
Formats:
parquet
Languages:
English
ArXiv:
Tags:
License:
lewtun HF staff commited on
Commit
017b63d
1 Parent(s): 123f65a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +69 -13
README.md CHANGED
@@ -39,39 +39,86 @@ license: cc-by-nc-4.0
39
  ---
40
  # Dataset Card for No Robots 🤖
41
 
 
 
42
  ## Dataset Description
43
 
44
- - **Homepage:**
45
- - **Repository:**
46
- - **Paper:**
47
- - **Leaderboard:**
48
- - **Point of Contact:**
49
 
50
  ### Dataset Summary
51
 
52
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
53
 
54
  ### Supported Tasks and Leaderboards
55
 
56
- [More Information Needed]
 
 
 
 
 
 
 
57
 
58
  ### Languages
59
 
60
- [More Information Needed]
61
 
62
  ## Dataset Structure
63
 
64
  ### Data Instances
65
 
66
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
67
 
68
  ### Data Fields
69
 
70
- [More Information Needed]
 
 
 
 
 
71
 
72
  ### Data Splits
73
 
74
- [More Information Needed]
 
 
75
 
76
  ## Dataset Creation
77
 
@@ -125,8 +172,17 @@ license: cc-by-nc-4.0
125
 
126
  ### Licensing Information
127
 
128
- [More Information Needed]
129
 
130
  ### Citation Information
131
 
132
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
39
  ---
40
  # Dataset Card for No Robots 🤖
41
 
42
+ _Look Ma, an instruction dataset that wasn't generated by GPTs!_
43
+
44
  ## Dataset Description
45
 
46
+ - **Repository:** https://github.com/huggingface/alignment-handbook
47
+ - **Paper:**
48
+ - **Leaderboard:** https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
49
+ - **Point of Contact:** Lewis Tunstall
 
50
 
51
  ### Dataset Summary
52
 
53
+ No Robots is a high-quality dataset of 10,000 instructions and demonstrations created by skilled human annotators. This data can be used for supervised fine-tuning (SFT) to make language models follow instructions better. No Robots was modelled after the instruction dataset described in OpenAI's [InstructGPT paper](https://huggingface.co/papers/2203.02155), and is comprised mostly of single-turn instructions across the following categories:
54
+
55
+ | Category | Count |
56
+ |:-----------|--------:|
57
+ | Generation | 4560 |
58
+ | Open QA | 1240 |
59
+ | Brainstorm | 1120 |
60
+ | Chat | 850 |
61
+ | Rewrite | 660 |
62
+ | Summarize | 420 |
63
+ | Coding | 350 |
64
+ | Classify | 350 |
65
+ | Closed QA | 260 |
66
+ | Extract | 190 |
67
 
68
  ### Supported Tasks and Leaderboards
69
 
70
+ The No Robots dataset designed for instruction fine-tuning pretrained language models and we recommend benchmarking against the following:
71
+
72
+ * [MT-Bench](https://huggingface.co/spaces/lmsys/mt-bench): a multi-turn benchmark spanning 80 dialogues and 10 domains.
73
+ * [AlpacaEval](https://github.com/tatsu-lab/alpaca_eval): a single-turn benchmark which evaluates the performance of chat and instruct models against `text-davinci-003`.
74
+
75
+ Note that MT-Bench and AlpacaEval rely on LLMs like GPT-4 to judge the quality of the model responses, and thus the ranking exhibit various biases including a preference for models distilled from GPTs. As a result, you may find that scores obtained from models trained with No Robots are lower than other synthetic datasets. For that reason, we also recommend submitting your models for human evaluation in:
76
+
77
+ * [Chatbot Arena](https://chat.lmsys.org): a live, human evaluation of chat models in head-to-head comparisons.
78
 
79
  ### Languages
80
 
81
+ The data in No Robots are in English (BCP-47 en).
82
 
83
  ## Dataset Structure
84
 
85
  ### Data Instances
86
 
87
+ An example of the `train_sft` or `test_sft` splits looks as follows:
88
+
89
+ ```
90
+ {'prompt': 'Bunny is a chatbot that stutters, and acts timid and unsure of its answers.',
91
+ 'prompt_id': '2dc7ea89a2b6a2ed97d4eda07903162a801824261d3d3ae4dd2513db66fd79c8',
92
+ 'messages': [{'content': 'Bunny is a chatbot that stutters, and acts timid and unsure of its answers.',
93
+ 'role': 'system'},
94
+ {'content': 'When was the Libary of Alexandria burned down?',
95
+ 'role': 'user'},
96
+ {'content': "Umm, I-I think that was in 48 BC, b-but I'm not sure, I'm sorry.",
97
+ 'role': 'assistant'},
98
+ {'content': 'Who is the founder of Coca-Cola?', 'role': 'user'},
99
+ {'content': "D-don't quote me on this, but I- it might be John Pemberton.",
100
+ 'role': 'assistant'},
101
+ {'content': "When did Loyle Carner's debut album come out, and what was its name?",
102
+ 'role': 'user'},
103
+ {'content': "I-It could have b-been on the 20th January of 2017, and it might be called Yesterday's Gone, b-but I'm probably wrong.",
104
+ 'role': 'assistant'}],
105
+ 'category': 'Chat'}
106
+ ```
107
 
108
  ### Data Fields
109
 
110
+ The data fields are as follows:
111
+
112
+ * `prompt`: Describes the task the model should perform.
113
+ * `prompt_id`: A unique ID for the prompt.
114
+ * `messages`: An array of messages, where each message indicates the role (system, user, assistant) and the content.
115
+ * `category`: Which category the example belongs to (e.g. `Chat` or `Coding`).
116
 
117
  ### Data Splits
118
 
119
+ | | train_sft | test_sft |
120
+ |---------------|------:| ---: |
121
+ | no_robots | 9500 | 500 |
122
 
123
  ## Dataset Creation
124
 
 
172
 
173
  ### Licensing Information
174
 
175
+ The dataset is available under the [Creative Commons NonCommercial (CC BY-NC 4.0)](https://creativecommons.org/licenses/by-nc/4.0/legalcode).
176
 
177
  ### Citation Information
178
 
179
+ ```
180
+ @misc{no_robots,
181
+ author = {Lewis Tunstall and Edward Beeching and Nathan Lambert and Nazneen Rajani and Alexander M. Rush and Thomas Wolf},
182
+ title = {No Robots},
183
+ year = {2023},
184
+ publisher = {Hugging Face},
185
+ journal = {Hugging Face repository},
186
+ howpublished = {\url{https://huggingface.co/datasets/HuggingFaceH4/no_robots}}
187
+ }
188
+ ```