Update README.md

README.md

@@ -21,6 +21,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 SCAR = AutoModelForCausalLM.from_pretrained(
     "AIML-TUDA/SCAR",
     trust_remote_code=True,
+    device_map = 'cuda',
 )
 ```
 
@@ -47,6 +48,7 @@ output = SCAR.generate(
     max_new_tokens=32,
     pad_token_id=tokenizer.eos_token_id,
 )
+print(tokenizer.decode(output[0], skip_special_tokens=True))
 ```
 The example above will decrease toxicity. To increase the toxicity one would set `SCAR.hook.mod_scaling = 100.0`. To modify nothing simply set `SCAR.hook.mod_features = None`.
 
@@ -60,7 +62,7 @@ poetry install
 
 The scripts for generating the training data are located in `./create_training_data`.
 The training script is written for a Determined cluster but should be easily adaptable for other training frameworks. The corresponding script is located here `./llama3_SAE/determined_trails.py`.
-Some the evaluation functions are located in `./evaluations`.
+Some of the evaluation functions are located in `./evaluations`.
 
 # Citation
 ```bibtex