Update README.md
Browse files
README.md
CHANGED
|
@@ -95,13 +95,12 @@ The model is designed for various domains, including:
|
|
| 95 |
|
| 96 |
---
|
| 97 |
|
| 98 |
-
##
|
| 99 |
-
|
| 100 |
-
The model excels in reasoning benchmarks:
|
| 101 |
-
- **ARC (AI2 Reasoning Challenge):** High accuracy in logical and commonsense tasks.
|
| 102 |
-
- **GSM8K (Math Reasoning):** Superior results in multi-step problem-solving.
|
| 103 |
-
- **CommonsenseQA:** Strong comprehension of everyday reasoning tasks.
|
| 104 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
---
|
| 106 |
|
| 107 |
## Ethical Considerations
|
|
|
|
| 95 |
|
| 96 |
---
|
| 97 |
|
| 98 |
+
## Limitations
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
|
| 100 |
+
- **Hallucinations** Model might hallucinate in very long context problems.
|
| 101 |
+
- **Unclosed tags** As the model gets deep into thinking and reflecting ,it has a tendency to not close thinking or critique tags .
|
| 102 |
+
- **Tags Compression** As the model gets confident in the answer , it will use less and less tags and might have everything in the <Thinking> Tag ,instead of reasoning and going step by step.
|
| 103 |
+
- **High Resource** This Model is Resource intensive and needs a lot of uninterrupted computing , since it's continuously generating tokens to reason , so it might work the best with consumer hardware.
|
| 104 |
---
|
| 105 |
|
| 106 |
## Ethical Considerations
|