|
|
--- |
|
|
license: apache-2.0 |
|
|
datasets: |
|
|
- Anthropic/hh-rlhf |
|
|
language: |
|
|
- en |
|
|
pipeline_tag: text-generation |
|
|
--- |
|
|
|
|
|
|
|
|
# GPT-2 Medium Fine-Tuned on Anthropic-hh Dataset |
|
|
|
|
|
This repository houses a GPT-2 Medium model fine-tuned on the Anthropic-hh dataset. The fine-tuning process involved masking Human's utterances, with the loss computed exclusively on the Assistant's responses. |
|
|
|
|
|
## Model Information |
|
|
|
|
|
- **Base Model:** GPT-2 Medium |
|
|
- **Training Data:** Anthropic-hh dataset |
|
|
- **Fine-Tuning Approach:** Supervised fine-tuning with a focus on Assistant's responses. |
|
|
|
|
|
## How to Use |
|
|
|
|
|
```python |
|
|
from transformers import GPT2LMHeadModel, GPT2Tokenizer |
|
|
|
|
|
# Load tokenizer and model |
|
|
tokenizer = GPT2Tokenizer.from_pretrained("RaushanTurganbay/GPT2_instruct_tuned") |
|
|
model = GPT2LMHeadModel.from_pretrained("RaushanTurganbay/GPT2_instruct_tuned") |
|
|
|
|
|
# Generate responses |
|
|
prompt = "Your input prompt here" |
|
|
inputs = tokenizer(prompt, return_tensors="pt") |
|
|
outputs = model.generate(**inputs, max_length=150, num_return_sequences=1) |
|
|
|
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
|
``` |
|
|
|
