Update README.md
Browse files
README.md
CHANGED
@@ -27,6 +27,14 @@ The GGUFs uploaded are full FP16 precision.
|
|
27 |
User: {prompt}
|
28 |
A: {response}
|
29 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
## Training Details:
|
31 |
- 1x RTX 3070 8GB
|
32 |
- 1x Ryzen 3 3700x
|
@@ -34,6 +42,11 @@ A: {response}
|
|
34 |
- Approx 50k tokens (>0.01 epoches)
|
35 |
- Training data = 1 billion tokens
|
36 |
|
|
|
|
|
|
|
|
|
|
|
37 |
## Example output:
|
38 |
```
|
39 |
User: Write a tutorial about caring for dogs in the style of WikiHow.
|
|
|
27 |
User: {prompt}
|
28 |
A: {response}
|
29 |
```
|
30 |
+
|
31 |
+
Please structure your prompts this way for maximum performance:
|
32 |
+
```
|
33 |
+
Write a long and very detailed tutorial on '{task}', in the style of WikiHow.
|
34 |
+
Include in depth explanations for each step and how it helps achieve the desired outcome, inluding key tips and guidelines.
|
35 |
+
Ensure clarity and practicality, allowing readers to easily follow and apply the instructions. Do not use images.
|
36 |
+
```
|
37 |
+
|
38 |
## Training Details:
|
39 |
- 1x RTX 3070 8GB
|
40 |
- 1x Ryzen 3 3700x
|
|
|
42 |
- Approx 50k tokens (>0.01 epoches)
|
43 |
- Training data = 1 billion tokens
|
44 |
|
45 |
+
## Notes:
|
46 |
+
|
47 |
+
The model is a bit biased, often talking about relationships and such without explicit prompting. This is likely due to the nature of WikiHow articles, because
|
48 |
+
most are about specific topics. I will attempt to reduce this bias by filtering the dataset, it should be ready in approx 2 weeks.
|
49 |
+
|
50 |
## Example output:
|
51 |
```
|
52 |
User: Write a tutorial about caring for dogs in the style of WikiHow.
|