You can deploy this model and start earning money today!

PaDaS-Lab/
SynStOp

Views

No views yet

Minimal dataset for intended for LM development and testing using python string operations. The dataset is created by running different one line python string operations on random strings The idea is, that transformer implementation can learn the string operations and that this task is a good proxy tasks for other transformer operations on real languages and real tasks. Consequently, the data set is small and can be used in the development process without large scale infrastructures.

There are different configurations for the data set.

small: contains below 50k instances of various string length and only contains slicing operations, i.e. all python operations expressable with s[i:j:s] (which also includes string reversal).
- you can further choose different subsets according to either length or the kind of operation
small10: like small, but only strings to length 10
small15: like small, but only strings to length 15
small20: like small, but only strings to length 20

The fields have the following meaning:

input: input string, i.e. the string and the string operation
output: output of the string operation
code: code for running the string operation in python,
res_var: name of the result variable
operation: kind of operation:
- step_x for s[::x]
- char_at_x for s[x]
- slice_x:y for s[x:y]
- slice_step_x:y:z for s[x:y:z]
- slice_reverse_i:j:k for s[i:i+j][::k]

Siblings of data contain additional metadata information about the dataset.

prompt describes possible prompts based on that data splitted into input prompts / output prompts

PaDaS-Lab/SynStOp

PaDaS-Lab/
SynStOp