Update README.md
Browse files
README.md
CHANGED
@@ -3,17 +3,20 @@ library_name: transformers
|
|
3 |
language:
|
4 |
- fi
|
5 |
---
|
|
|
6 |
|
7 |
-
# Model Card for Model ID
|
8 |
|
9 |
-
|
10 |
-
|
|
|
|
|
11 |
LIMA from https://github.com/TurkuNLP/finnish-instructions
|
12 |
Dolly from https://github.com/TurkuNLP/finnish-instructions
|
13 |
OASST from https://github.com/TurkuNLP/finnish-instructions
|
14 |
Heavily filtered version of Ultrachat + deepl translations from https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized/viewer/default/train_sft
|
15 |
|
16 |
|
|
|
17 |
### How to use
|
18 |
|
19 |
Here is an example of using this model with Unsloth with some generation arguments you can modify:
|
@@ -123,9 +126,9 @@ VASTAUS:
|
|
123 |
|
124 |
### Limitations and bias
|
125 |
|
126 |
-
The training data used for this model contains a lot of content from the internet, which is far from neutral.
|
127 |
-
|
128 |
-
To reduce toxic content,
|
129 |
|
130 |
### Finetuning
|
131 |
|
|
|
3 |
language:
|
4 |
- fi
|
5 |
---
|
6 |
+
## Model description
|
7 |
|
|
|
8 |
|
9 |
+
This is an early v0.1 version release of our Instruct finetuned model from https://huggingface.co/Finnish-NLP/llama-7b-finnish
|
10 |
+
Model was trained for 2 epochs using 11014 samples and for this release we chose checkpoint at 2500/4048 steps.
|
11 |
+
|
12 |
+
For finetuning we used mix of following datasources:
|
13 |
LIMA from https://github.com/TurkuNLP/finnish-instructions
|
14 |
Dolly from https://github.com/TurkuNLP/finnish-instructions
|
15 |
OASST from https://github.com/TurkuNLP/finnish-instructions
|
16 |
Heavily filtered version of Ultrachat + deepl translations from https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized/viewer/default/train_sft
|
17 |
|
18 |
|
19 |
+
|
20 |
### How to use
|
21 |
|
22 |
Here is an example of using this model with Unsloth with some generation arguments you can modify:
|
|
|
126 |
|
127 |
### Limitations and bias
|
128 |
|
129 |
+
The training data used for this model contains a lot of content from the internet, which is far from neutral.
|
130 |
+
Therefore, the model can have biased predictions. This bias will also affect all fine-tuned versions of this model.
|
131 |
+
To reduce toxic content, the pretrained version of thiis model was trained with dataset filtered with a toxicity classifier but it cannot truly eliminate all toxic text.
|
132 |
|
133 |
### Finetuning
|
134 |
|