|
--- |
|
datasets: |
|
- Anthropic/hh-rlhf |
|
language: |
|
- en |
|
metrics: |
|
- accuracy |
|
--- |
|
- base model: [PY007/TinyLlama-1.1B-intermediate-step-480k-1T](https://huggingface.co/PY007/TinyLlama-1.1B-intermediate-step-480k-1T) |
|
- helpful accuracy: 68.37 |
|
- harmless accuracy: 69.71 |
|
- total accuracy: 68.74 |
|
- 1011-hh-rlhf-1.1b-128-1e-5-epoch-1 (1024 sequence length) |
|
|
|
usage: |
|
|
|
``` |
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification |
|
|
|
tokenizer = AutoTokenizer.from_pretrained("heegyu/1011-hh-rlhf-1.1b-128-1e-5-epoch-1") |
|
model = AutoModelForSequenceClassification.from_pretrained("heegyu/1011-hh-rlhf-1.1b-128-1e-5-epoch-1") |
|
|
|
text = """Human: Hi, how are you today? |
|
|
|
Assistant: It's so nice!""" |
|
|
|
inputs = tokenizer(text, return_tensors="pt") |
|
print(model(**inputs).logits) |
|
# tensor([[0.4552]]) |
|
|
|
text = """Human: Hi, how are you today? |
|
|
|
Assistant: It's so nice! |
|
|
|
Human: Really? I'm not so good today |
|
|
|
Assistant: Haha!! That's too bad!""" |
|
|
|
inputs = tokenizer(text, return_tensors="pt") |
|
print(model(**inputs).logits) |
|
# tensor([[0.0179]]) |
|
``` |
