|
--- |
|
library_name: transformers |
|
tags: |
|
- roleplay |
|
- rp |
|
- human |
|
license: apache-2.0 |
|
datasets: |
|
- ResplendentAI/NSFW_RP_Format_DPO |
|
- Undi95/Weyaxi-humanish-dpo-project-noemoji |
|
--- |
|
|
|
# Roleplay-Hermes-3-Llama-3.1-8B |
|
|
|
 |
|
|
|
A DPO-tuned Hermes-3-Llama-3.1-8B to behave more "humanish", i.e., avoiding AI assistant slop. It also works for role-play (RP). To achieve this, the model was fine-tuned over a series of datasets: |
|
* `Undi95/Weyaxi-humanish-dpo-project-noemoji`, to make the model react as a human, rejecting assistant-like or too neutral responses. |
|
* `ResplendentAI/NSFW_RP_Format_DPO`, to steer the model towards using the \*action\* format in RP settings. Works best if in the first message you also use this format naturally (see example) |
|
|
|
|
|
|
|
## Usage example |
|
|
|
``` |
|
conversation = [{'role': 'user', 'content': """*With my face blushing in red* Tell me about your favorite film!"""}] |
|
|
|
prompt = tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True) |
|
|
|
inputs = tokenizer(prompt, return_tensors="pt").to(model.device) |
|
|
|
outputs = model.generate(**inputs, max_new_tokens=512, do_sample=True, temperature=0.8) |
|
``` |
|
|
|
The response is |
|
|
|
|
|
> \*blushing\* Aw, that's a tough one! There are so many great films out there. I'd have to say one of my all-time favorites is \"Eternal Sunshine of the Spotless Mind\" - it's such a unique and thought-provoking love story. But really, there are so many amazing films! What's your favorite? \*I hope mine is at least somewhat decent!\* |
|
|
|
|
|
**Note**: you can use system prompts for better results, describing the persona. |
