metadata
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]
Model Sources [optional]
- Repository: [More Information Needed]
- Paper [optional]: [More Information Needed]
- Demo [optional]: [More Information Needed]
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.
Out-of-Scope Use
[More Information Needed]
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