|
|
--- |
|
|
license: llama3.3 |
|
|
base_model: meta-llama/Llama-3.3-70B-Instruct |
|
|
tags: |
|
|
- lora |
|
|
- llama |
|
|
- character-simulation |
|
|
- political-persona |
|
|
language: |
|
|
- en |
|
|
- ru |
|
|
--- |
|
|
|
|
|
# Putin Bot - LoRA Adapter |
|
|
|
|
|
LoRA fine-tuning weights for Llama 3.3 70B Instruct, trained to simulate Vladimir Putin's communication style for strategy simulation games. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Base Model**: meta-llama/Llama-3.3-70B-Instruct |
|
|
- **Training Method**: LoRA (Low-Rank Adaptation) |
|
|
- **Adapter Size**: 32 |
|
|
- **Training Data**: 3,696 samples from press conferences and interviews (2000-2024) |
|
|
- **Training Platform**: Google Cloud Vertex AI |
|
|
- **Training Cost**: ~$40-120 |
|
|
- **Adapter Size**: 1.5GB |
|
|
|
|
|
## Training Data Sources |
|
|
|
|
|
- Kremlin press conference transcripts (575 documents) |
|
|
- Tucker Carlson interview (Feb 2024) |
|
|
- Oliver Stone interviews (2017) |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
from peft import PeftModel |
|
|
|
|
|
# Load base model |
|
|
base_model = "meta-llama/Llama-3.3-70B-Instruct" |
|
|
model = AutoModelForCausalLM.from_pretrained( |
|
|
base_model, |
|
|
torch_dtype="auto", |
|
|
device_map="auto" |
|
|
) |
|
|
tokenizer = AutoTokenizer.from_pretrained(base_model) |
|
|
|
|
|
# Load LoRA adapter |
|
|
model = PeftModel.from_pretrained(model, "kennethpayne01/putin-bot-lora") |
|
|
|
|
|
# Generate |
|
|
messages = [ |
|
|
{"role": "system", "content": "You are Vladimir Putin, President of Russia."}, |
|
|
{"role": "user", "content": "What is your view on NATO expansion?"} |
|
|
] |
|
|
|
|
|
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True) |
|
|
outputs = model.generate(inputs, max_new_tokens=512, temperature=0.7) |
|
|
response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True) |
|
|
print(response) |
|
|
``` |
|
|
|
|
|
## Hardware Requirements |
|
|
|
|
|
- **Full precision**: ~140GB VRAM (2-4x A100 80GB) |
|
|
- **4-bit quantization**: ~40GB VRAM (1x A100 80GB) |
|
|
- **8-bit quantization**: ~70GB VRAM (1x A100 80GB) |
|
|
|
|
|
## Training Configuration |
|
|
|
|
|
- Learning rate: 0.0001 |
|
|
- Epochs: 3 |
|
|
- LoRA rank (r): 32 |
|
|
- LoRA alpha: 64 |
|
|
- Target modules: All attention layers |
|
|
- Training time: ~2-4 hours |
|
|
- Platform: Vertex AI with A100 GPUs |
|
|
|
|
|
## Limitations |
|
|
|
|
|
- Model trained on public statements; may not reflect private views |
|
|
- Data range 2000-2024; current events after Dec 2024 not included |
|
|
- English translations may lose nuance from original Russian |
|
|
- Designed for simulation/entertainment, not policy analysis |
|
|
|
|
|
## Ethical Considerations |
|
|
|
|
|
This model is created for strategy simulation games and educational purposes. It should not be used to: |
|
|
- Spread misinformation or propaganda |
|
|
- Impersonate real individuals for deception |
|
|
- Generate harmful or misleading content |
|
|
|
|
|
## License |
|
|
|
|
|
This adapter is released under the Llama 3.3 license. See base model license for details. |
|
|
|
|
|
## Citation |
|
|
|
|
|
```bibtex |
|
|
@misc{putin-bot-lora, |
|
|
author = {Your Name}, |
|
|
title = {Putin Bot LoRA Adapter}, |
|
|
year = {2025}, |
|
|
publisher = {Hugging Face}, |
|
|
howpublished = {\url{https://huggingface.co/kennethpayne01/putin-bot-lora}} |
|
|
} |
|
|
``` |
|
|
|
|
|
## Repository |
|
|
|
|
|
Full code and training pipeline: [GitHub Repository](#) |
|
|
|
