jcarbonnell commited on
Commit
a1c006f
1 Parent(s): 2c45f1a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +7 -26
README.md CHANGED
@@ -1,26 +1,7 @@
1
- ---
2
- dataset_info:
3
- features:
4
- - name: repoName
5
- dtype: string
6
- - name: tree
7
- dtype: string
8
- - name: readme
9
- dtype: string
10
- splits:
11
- - name: train
12
- num_bytes: 142452923
13
- num_examples: 2863
14
- - name: val
15
- num_bytes: 18136850
16
- num_examples: 319
17
- download_size: 62341533
18
- dataset_size: 160589773
19
- configs:
20
- - config_name: default
21
- data_files:
22
- - split: train
23
- path: data/train-*
24
- - split: val
25
- path: data/val-*
26
- ---
 
1
+ The structTuningNEAR dataset is a subset of the original nearData dataset, specially prepared for the structure-aware finetuning of a pre-trained LLM.
2
+
3
+ The Structure-Aware Finetuning approach instructs the model with dApps trees and their corresponding readme files. It aim to give the model a good knowledge of the whole dApp logic so that when a user asks it to create an app, the model will primarily provide an output focused on the big-picture structure and its description. The goal of Structure-Aware Finetuning is to bypass the limited logic of the 'next-token prediction', which sometimes spins the model in 'dumb loops' while iterating over complex coding challenges. Structure-aware code LLMs should also be of great use for code understanding and code discussion.
4
+
5
+ The structTuningNEAR dataset is made of:
6
+ - nearDappsTrees: 3414 text files representing the tree structure extracted from the nearDapps files.
7
+ - nearDappsReadme: 23166 readme files extracted in text formats from the nearDapps files.