tokenizer and torch variables fixed
Browse files
README.md
CHANGED
@@ -41,33 +41,34 @@ base_model:
|
|
41 |
### Direct Use
|
42 |
|
43 |
```python
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
|
|
71 |
```
|
72 |
|
73 |
## Evaluation
|
|
|
41 |
### Direct Use
|
42 |
|
43 |
```python
|
44 |
+
from transformers import AutoTokenizer, pipeline
|
45 |
+
import torch
|
46 |
+
|
47 |
+
model_name = "NCSOFT/Llama-3-OffsetBias-RM-8B"
|
48 |
+
rm_tokenizer = AutoTokenizer.from_pretrained(model_name)
|
49 |
+
rm_pipe = pipeline(
|
50 |
+
"sentiment-analysis",
|
51 |
+
model=model_name,
|
52 |
+
device="auto",
|
53 |
+
tokenizer=rm_tokenizer,
|
54 |
+
model_kwargs={"torch_dtype": torch.bfloat16}
|
55 |
+
)
|
56 |
+
|
57 |
+
pipe_kwargs = {
|
58 |
+
"return_all_scores": True,
|
59 |
+
"function_to_apply": "none",
|
60 |
+
"batch_size": 1
|
61 |
+
}
|
62 |
+
|
63 |
+
chat = [
|
64 |
+
{"role": "user", "content": "Hello, how are you?"},
|
65 |
+
{"role": "assistant", "content": "I'm doing great. How can I help you today?"},
|
66 |
+
{"role": "user", "content": "I'd like to show off how chat templating works!"},
|
67 |
+
]
|
68 |
+
|
69 |
+
test_texts = [rm_tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=False).replace(rm_tokenizer.bos_token, "")]
|
70 |
+
pipe_outputs = rm_pipe(test_texts, **pipe_kwargs)
|
71 |
+
rewards = [output[0]["score"] for output in pipe_outputs]
|
72 |
```
|
73 |
|
74 |
## Evaluation
|