File size: 1,467 Bytes
7280b63 7247118 17aaa37 7247118 17aaa37 7280b63 7247118 17aaa37 7247118 17aaa37 7280b63 7247118 7280b63 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 |
---
base_model: AI-Sweden-Models/gpt-sw3-6.7b-v2-instruct
language:
- sv
- da
- 'no'
- en
pipeline_tag: text-generation
tags:
- translation
---
# Model Card for gpt-sw3-6.7b-v2-translator-gguf
The `gpt-sw3-6.7b-v2-translator` is a finetuned version of `gpt-sw3-6.7b-v2-instruct` on a carefully selected translation pair dataset that was gathered by AI Sweden.
## Intended usage:
Translate text data from English to Swedish, or Swedish to English.
## How to use:
Translate from English to Swedish:
```bash
FROM ./gpt-sw3-6-7b-v2-translator-SIZE.gguf
TEMPLATE "<|endoftext|><s>User: Översätt till Svenska från Engelska\n{{ .Prompt }}<s>Bot:"
PARAMETER stop <s>
```
Translate from Swedish to English:
```bash
FROM ./gpt-sw3-6-7b-v2-translator-SIZE.gguf
TEMPLATE "<|endoftext|><s>User: Översätt till Engelska från Svenska\n{{ .Prompt }}<s>Bot:"
PARAMETER stop <s>
```
## Versions:
```
gpt-sw3-6-7b-v2-translator-q4.gguf
gpt-sw3-6-7b-v2-translator-q8.gguf
gpt-sw3-6-7b-v2-translator-f16.gguf
gpt-sw3-6-7b-v2-translator-f32.gguf
```
## Training & Data:
The training was done on 1 NVIDIA DGX using DeepSpeed ZeRO 3 for three epochs on roughly 4GB of carefully selected translation data. It is a full finetune of all of the model parameters.
| Epoch | Training Loss | Evaluation Loss |
|-------|---------------|-----------------|
| 1 | 1.309 | 1.281 |
| 2 | 1.161 | 1.242 |
| 3 | 1.053 | 1.219 | |