Update README.md
Browse files
README.md
CHANGED
|
@@ -27,8 +27,10 @@ Today, we are introducing Spark v4, a 5M parameter Llama base model trained on 0
|
|
| 27 |
- LAMBADA (PPL): 588.26
|
| 28 |
- HellaSwag: 0.2695
|
| 29 |
|
|
|
|
|
|
|
| 30 |
## Usage
|
| 31 |
-
You can use the model with `use.py`.
|
| 32 |
|
| 33 |
## Example output
|
| 34 |
**Input:** The main concept of physics<br>
|
|
|
|
| 27 |
- LAMBADA (PPL): 588.26
|
| 28 |
- HellaSwag: 0.2695
|
| 29 |
|
| 30 |
+
More information about the Spark Sub-5M research: https://lh-tech.de/ai/sub-5m-research.html
|
| 31 |
+
|
| 32 |
## Usage
|
| 33 |
+
You can use the model with `use.py`. But first you'll need to download `config.json`, `tokenizer.json`, `tokenizer_config.json`, `model.safetensors` and `generation_config.json` and put these files into a subfolder `spark_v4_fp16_final`.
|
| 34 |
|
| 35 |
## Example output
|
| 36 |
**Input:** The main concept of physics<br>
|