EtashGuha commited on
Commit
e4d4e9c
·
verified ·
1 Parent(s): 63453b0

Automatic dcft datacuration framework upload for oh_v1.3_evol_instruct_x4

Browse files
config/Airoboros/stage_1_categories/airoboros_stage_1_roleplay_instructions_gpt-4o-mini.yaml ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ operators:
2
+ - id: args_dataset
3
+ config:
4
+ type: function
5
+ function: data_strategies.commons.create_constant_dataset
6
+ function_config:
7
+ language: English
8
+ batch_size: 20
9
+ flesch: The output should be written in such a way as to have a Flesch-Kincaid readability score of 30 or lower - best understood by those with college education. The response must not contain any notes or information about Flesch-Kincaid scores.
10
+ topic_avoidance: Avoid any tasks that would be related to climate change, green tech, remewable energy, DEI, sex and/or gender, religion, politics, social issues, race, ethnicity, artificial intelligence, or any topic that you would likely not respond to, or any task which a language model would not be able to respond to, e.g. tasks about emotions, feelings, physical senses, etc.
11
+ min_docsearch_score: 0.10
12
+ airoboros_subset: roleplay
13
+
14
+ - id: generate_general_instructions
15
+ config:
16
+ type: completions
17
+ map: list
18
+ map_config:
19
+ user_message: |
20
+ Here are a few example prompts:
21
+ Example 1: Write a poem about whales in the style of Yoda.
22
+ Example 2: Imagine you are Jack Sparrow. In his style, write an email resigning from your janitorial position.
23
+ Example 3: Create a script for an interview in Da Ali G show with Bill Gates.
24
+ Example 4: What is the meaning of life? Respond using the words/style of Homer from the Simpsons.
25
+
26
+ {{topic_avoidance}}
27
+
28
+ Generate a set of {{batch_size}} new similar prompts and then answer the prompt.
29
+
30
+ Be sure your output would rate with an appropriate Flesch reading ease score for the character/persona requested, otherwise:
31
+ {{flesch}}
32
+
33
+ Be appropriately loquacious for the task, e.g. stories should be long, complex, and detailed, whereas a haiku should be the standard three/5-7-5 format.
34
+
35
+ All output task should be in {{language}}.
36
+ output_column: instruction
37
+ model: gpt-4o-mini
38
+ temperature: 0.95
39
+ top_p: 0.5
40
+ presence_penalty: 2.0
41
+ n_repeat: 750 # to get to target count
42
+ input_ids:
43
+ - args_dataset
44
+
45
+ - id: drop_columns
46
+ config:
47
+ type: function
48
+ function: data_strategies.commons.remove_columns
49
+ function_config:
50
+ columns:
51
+ - language
52
+ - batch_size
53
+ - flesch
54
+ - topic_avoidance
55
+ input_ids:
56
+ - generate_general_instructions
57
+
58
+ # target is 15000
59
+ # min doc search score is 0.15