|
--- |
|
license: mit |
|
datasets: pt-sk/toxic_classification |
|
tags: ["PPO", "RLHF"] |
|
--- |
|
Aligning the model using Proximal Policy Optimization (PPO). The goal is to train the model to generate non-toxic reviews. The training process utilizes the `trl` library for reinforcement learning, the `transformers` library for model handling, and `datasets` for dataset management. |
|
Implementation code is available here: [GitHub](https://github.com/sathishkumar67/GPT-2-Non-Toxic-RLHF) |
|
```python |
|
# Load model and tokenizer directly |
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
tokenizer = AutoTokenizer.from_pretrained("pt-sk/GPT2_NonToxic") |
|
model = AutoModelForCausalLM.from_pretrained("pt-sk/GPT2_NonToxic") |
|
|
|
# Example usage |
|
input_text = "The movie was fantastic" |
|
inputs = tokenizer(input_text, return_tensors='pt') |
|
outputs = model.generate(**inputs) |
|
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
``` |
