neurallambda
commited on
Commit
•
f2522dd
1
Parent(s):
1553fae
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -1,34 +1,91 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
# Arithmetic Puzzles Dataset
|
3 |
+
|
4 |
+
A collection of arithmetic puzzles with heavy use of variable assignment. Current LLMs struggle with variable indirection/multi-hop reasoning, this should be a tough test for them.
|
5 |
+
|
6 |
+
Inputs are a list of strings representing variable assignments (`c=a+b`), and the output is the integer answer.
|
7 |
+
|
8 |
+
Outputs are filtered to be between [-100, 100], and self-reference/looped dependencies are forbidden.
|
9 |
+
|
10 |
+
Splits include:
|
11 |
+
|
12 |
+
- `train_small/test_small` which includes 10k total examples of puzzles with up to 10 variables.
|
13 |
+
- `train_large/test_large` which includes 10k total examples of puzzles with up to 100 variables.
|
14 |
+
|
15 |
+
Conceptually the data looks like this:
|
16 |
+
|
17 |
+
```python
|
18 |
+
Input:
|
19 |
+
|
20 |
+
a=1
|
21 |
+
b=2
|
22 |
+
c=a+b
|
23 |
+
solve(c)=
|
24 |
+
|
25 |
+
Output:
|
26 |
+
3
|
27 |
+
```
|
28 |
+
|
29 |
+
In actuality it looks like this:
|
30 |
+
|
31 |
+
```python
|
32 |
+
{
|
33 |
+
"input": ['var_0=1', 'var_1=2', 'var_2=a+b', 'solve(var_2)='],
|
34 |
+
"output": 3
|
35 |
+
}
|
36 |
+
```
|
37 |
+
|
38 |
+
|
39 |
+
### Loading the Dataset
|
40 |
+
|
41 |
+
```python
|
42 |
+
from datasets import load_dataset
|
43 |
+
|
44 |
+
# Load the entire dataset
|
45 |
+
dataset = load_dataset("neurallambda/arithmetic_puzzles")
|
46 |
+
|
47 |
+
# Load specific splits
|
48 |
+
train_small = load_dataset("neurallambda/arithmetic_puzzles", split="train_small")
|
49 |
+
test_small = load_dataset("neurallambda/arithmetic_puzzles", split="test_small")
|
50 |
+
train_large = load_dataset("neurallambda/arithmetic_puzzles", split="train_large")
|
51 |
+
test_large = load_dataset("neurallambda/arithmetic_puzzles", split="test_large")
|
52 |
+
```
|
53 |
+
|
54 |
+
### Preparing Inputs
|
55 |
+
|
56 |
+
To prepare the inputs as concatenated strings, you can do this:
|
57 |
+
|
58 |
+
```python
|
59 |
+
def prepare_input(example):
|
60 |
+
return {
|
61 |
+
"input_text": "
|
62 |
+
".join(example["input"]),
|
63 |
+
"output": example["output"]
|
64 |
+
}
|
65 |
+
|
66 |
+
# Apply the preparation to a specific split
|
67 |
+
train_small_prepared = train_small.map(prepare_input)
|
68 |
+
|
69 |
+
# Example of using the prepared dataset
|
70 |
+
for example in train_small_prepared.select(range(5)): # Show first 5 examples
|
71 |
+
print("Input:", example["input_text"])
|
72 |
+
print("Output:", example["output"])
|
73 |
+
print()
|
74 |
+
```
|
75 |
+
|
76 |
+
This will produce output similar to:
|
77 |
+
|
78 |
+
```
|
79 |
+
Input: var_0=5
|
80 |
+
var_1=2
|
81 |
+
var_2=-2 + -8
|
82 |
+
var_3=3
|
83 |
+
var_4=4
|
84 |
+
var_5=var_2
|
85 |
+
var_6=var_3 * 10
|
86 |
+
var_7=var_2 - var_0
|
87 |
+
var_8=var_1
|
88 |
+
var_9=-2 - 9
|
89 |
+
solve(var_3)=
|
90 |
+
Output: 3
|
91 |
+
```
|