Update README.md
Browse files
README.md
CHANGED
|
@@ -19,13 +19,20 @@ Anna was trained on German text with a similar methodology to [Boris](https://hu
|
|
| 19 |
# How to run
|
| 20 |
|
| 21 |
## Loading the model
|
|
|
|
| 22 |
```
|
| 23 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 24 |
|
| 25 |
tokenizer = AutoTokenizer.from_pretrained("Cedille/de-anna")
|
| 26 |
model = AutoModelForCausalLM.from_pretrained("Cedille/de-anna")
|
| 27 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
|
|
|
|
|
|
|
| 29 |
|
| 30 |
## Contact us
|
| 31 |
For any custom development please contact us at hello@cedille.ai.
|
|
|
|
| 19 |
# How to run
|
| 20 |
|
| 21 |
## Loading the model
|
| 22 |
+
### Base (requires 48+ GB of RAM)
|
| 23 |
```
|
| 24 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 25 |
|
| 26 |
tokenizer = AutoTokenizer.from_pretrained("Cedille/de-anna")
|
| 27 |
model = AutoModelForCausalLM.from_pretrained("Cedille/de-anna")
|
| 28 |
```
|
| 29 |
+
### Lower memory usage (loads on 16GB of RAM)
|
| 30 |
+
GPT_J models (link) have a parameter to only be loaded once ...
|
| 31 |
+
Combine that with half precision (fp16) ...
|
| 32 |
+
TO DO
|
| 33 |
|
| 34 |
+
## Generation
|
| 35 |
+
TO DO
|
| 36 |
|
| 37 |
## Contact us
|
| 38 |
For any custom development please contact us at hello@cedille.ai.
|