Improve things a bit
Browse files
README.md
CHANGED
|
@@ -19,7 +19,9 @@ Krikri is built on top of [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama
|
|
| 19 |
|
| 20 |
|
| 21 |
|
| 22 |
-
#
|
|
|
|
|
|
|
| 23 |
|
| 24 |
- Vocabulary extension of the Llama-3.1 tokenizer with Greek tokens
|
| 25 |
- 128k context length (approximately 80,000 Greek words)
|
|
@@ -41,13 +43,14 @@ Krikri is built on top of [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama
|
|
| 41 |
|
| 42 |
Chosen subsets of the 91 billion corpus were upsampled resulting in a size of **110 billion tokens**.
|
| 43 |
|
| 44 |
-
|
| 45 |
|
| 46 |
-
🚨 **More information
|
| 47 |
|
| 48 |
|
| 49 |
# How to use
|
| 50 |
|
|
|
|
| 51 |
|
| 52 |
```python
|
| 53 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
@@ -73,7 +76,7 @@ outputs = model.generate(input_prompt['input_ids'], max_new_tokens=256, do_sampl
|
|
| 73 |
print(tokenizer.batch_decode(outputs)[0])
|
| 74 |
```
|
| 75 |
|
| 76 |
-
|
| 77 |
|
| 78 |
```bash
|
| 79 |
vllm serve ilsp/Llama-Krikri-8B-Instruct \
|
|
|
|
| 19 |
|
| 20 |
|
| 21 |
|
| 22 |
+
# Model Information
|
| 23 |
+
|
| 24 |
+
## Base Model
|
| 25 |
|
| 26 |
- Vocabulary extension of the Llama-3.1 tokenizer with Greek tokens
|
| 27 |
- 128k context length (approximately 80,000 Greek words)
|
|
|
|
| 43 |
|
| 44 |
Chosen subsets of the 91 billion corpus were upsampled resulting in a size of **110 billion tokens**.
|
| 45 |
|
| 46 |
+
## Instruct Model
|
| 47 |
|
| 48 |
+
🚨 **More information on the post-training corpus and methdology coming soon.** 🚨
|
| 49 |
|
| 50 |
|
| 51 |
# How to use
|
| 52 |
|
| 53 |
+
## With Transformers
|
| 54 |
|
| 55 |
```python
|
| 56 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
| 76 |
print(tokenizer.batch_decode(outputs)[0])
|
| 77 |
```
|
| 78 |
|
| 79 |
+
## With OpenAI compatible server via vLLM
|
| 80 |
|
| 81 |
```bash
|
| 82 |
vllm serve ilsp/Llama-Krikri-8B-Instruct \
|