Update README.md
Browse files
README.md
CHANGED
|
@@ -6,6 +6,9 @@ pipeline_tag: text-generation
|
|
| 6 |
tags:
|
| 7 |
- pretrained
|
| 8 |
---
|
|
|
|
|
|
|
|
|
|
| 9 |
## Model Summary
|
| 10 |
|
| 11 |
NextTerm-47M is a pretrained transformer w/ 47.2M parameters, trained on 1.9 billion tokens of augmented data from the On-Line Encyclopedia of Integer Sequences (OEIS). It is designed to predict the next term in integer sequences. It displays exceptional in-context learning capabilities, and outperforms far larger generic LLMs on OEIS sequence completion tasks. It supports MLX and HuggingFace transformers.
|
|
@@ -32,8 +35,6 @@ Not all predictions are successful; example failure case below:
|
|
| 32 |
|
| 33 |
## Evaluation Results
|
| 34 |
|
| 35 |
-
[](radar_chart.png)
|
| 36 |
-
|
| 37 |
### Arithmetic Evaluation
|
| 38 |
|
| 39 |
The arithmetic evaluation consists of predicting the next term in sequences generated by polynomial functions of varying degrees (arithmetic, quadratic, cubic, quartic), across varying shot counts. The models are evaluated based on the accuracy of their predictions, w/ exact-match. NextTerm-47M outperforms all Qwen models <4B, though larger Qwen models do better on lower-degree polynomials.
|
|
|
|
| 6 |
tags:
|
| 7 |
- pretrained
|
| 8 |
---
|
| 9 |
+
|
| 10 |
+
[](radar_chart.png)
|
| 11 |
+
|
| 12 |
## Model Summary
|
| 13 |
|
| 14 |
NextTerm-47M is a pretrained transformer w/ 47.2M parameters, trained on 1.9 billion tokens of augmented data from the On-Line Encyclopedia of Integer Sequences (OEIS). It is designed to predict the next term in integer sequences. It displays exceptional in-context learning capabilities, and outperforms far larger generic LLMs on OEIS sequence completion tasks. It supports MLX and HuggingFace transformers.
|
|
|
|
| 35 |
|
| 36 |
## Evaluation Results
|
| 37 |
|
|
|
|
|
|
|
| 38 |
### Arithmetic Evaluation
|
| 39 |
|
| 40 |
The arithmetic evaluation consists of predicting the next term in sequences generated by polynomial functions of varying degrees (arithmetic, quadratic, cubic, quartic), across varying shot counts. The models are evaluated based on the accuracy of their predictions, w/ exact-match. NextTerm-47M outperforms all Qwen models <4B, though larger Qwen models do better on lower-degree polynomials.
|