Update README.md
Browse files
README.md
CHANGED
|
@@ -34,6 +34,8 @@ This model is a **fine-tuned LLaMA-based** (or derivative) model designed to (1)
|
|
| 34 |
|
| 35 |
## Training Approach
|
| 36 |
|
|
|
|
|
|
|
| 37 |
1. **Data Preprocessing**: We unified suicidal posts as `"suicidal"` and non-suicidal posts as `"non-suicidal"`.
|
| 38 |
2. **Multi-Task Instruction**: We used short prompts for classification tasks, and Q&A style prompts for therapy.
|
| 39 |
3. **Oversampling**: To ensure the model doesn’t just classify everything as “suicidal,” we oversampled the therapy data.
|
|
|
|
| 34 |
|
| 35 |
## Training Approach
|
| 36 |
|
| 37 |
+
To see a detailed Notebook of our training approach: https://colab.research.google.com/drive/1vbio7VWmkpQoTnDUg32TABxxf4VcBBeY?usp=sharing
|
| 38 |
+
|
| 39 |
1. **Data Preprocessing**: We unified suicidal posts as `"suicidal"` and non-suicidal posts as `"non-suicidal"`.
|
| 40 |
2. **Multi-Task Instruction**: We used short prompts for classification tasks, and Q&A style prompts for therapy.
|
| 41 |
3. **Oversampling**: To ensure the model doesn’t just classify everything as “suicidal,” we oversampled the therapy data.
|