Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,17 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- ihalage/sinhala-instruction-finetune-large
|
| 5 |
+
language:
|
| 6 |
+
- si
|
| 7 |
+
- en
|
| 8 |
+
---
|
| 9 |
+
[](https://github.com/ihalage/llama3-sinhala)
|
| 10 |
+
|
| 11 |
+
LLaMA3 (8B) model instruction finetuned to understand and respond in Sinhala language. `meta-llama/Meta-Llama-3-8B-Instruct` is finetuned on a reletively large dataset in
|
| 12 |
+
Sinhala compiled by translating English datasets such as ELI5 and Alpaca. The dataset in hosted in Hugging Face Datasets hub [(`sinhala-instruction-finetune-large`)](https://huggingface.co/datasets/ihalage/sinhala-instruction-finetune-large)
|
| 13 |
+
|
| 14 |
+
The original model is 4-bit quantized and finetuned with a causal language modelling (CLM) objective by adding LoRA adapters with a rank of 16 and a scaling factor of 32.
|
| 15 |
+
|
| 16 |
+
The finetuned `llama3-sinhala` model generates better responses in Sinhala compared to the original instruction finetuned model released by Meta.
|
| 17 |
+
See the github repo [llama3-sinhala](https://github.com/ihalage/llama3-sinhala) for more details.
|