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 |