File size: 1,282 Bytes
14427ae
d0c5ecd
 
d1e0c63
 
 
 
 
 
 
 
14427ae
d0c5ecd
 
 
d1e0c63
d0c5ecd
 
 
 
 
 
 
 
d1e0c63
 
 
 
 
 
 
d0c5ecd
 
 
d1e0c63
d0c5ecd
 
 
 
d1e0c63
d0c5ecd
 
 
 
d1e0c63
 
d0c5ecd
 
 
d1e0c63
d0c5ecd
d1e0c63
d0c5ecd
 
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
50
51
52
53
54
---
library_name: peft
base_model: NousResearch/Llama-2-7b-hf
license: mit
datasets:
- BiniyamAjaw/amharic_dataset_v2
language:
- am
metrics:
- bleu
pipeline_tag: text-generation
---

# Model Card for Model ID

Model fine tuned with LoRA on an Amharic Corpus of data collected from public telegram channels and groups.



## Model Details

### Model Description


- **Developed by:** [Biniyam Ajaw, Elias Assamnew]
- **Funded by:** [10 Academy]
- **Shared by [optional]:** [Biniyam Ajaw]
- **Model type:** [Text Generation]
- **Language(s) (NLP):** [Amharic - English]
- **License:** [MIT]
- **Finetuned from model [optional]:** [NousResearch-Llama2-7B-hf]

## Uses

The model is still in development and significantly lacks training data so it might not generate contents the way you want it to.


### Downstream Use [optional]

You can fine tune this model on labeled data for a specific domain. To get more pleasing results.


## Bias, Risks, and Limitations

The model is highly biased towards generating news content.
The model might repeat specific words because it is trained on a cleaned but unfiltered data because of the lack of tokens.



### Recommendations

The model is better of if you train it on labeled data if you want it to generate a content.

- PEFT 0.7.2.dev0