This LoRA adapter for Mistral-7B-Instruct-v0.2 has been fine-tuned to emulate Joe Biden's distinctive speaking style, discourse patterns, and policy positions. The model captures the measured cadence, personal anecdotes, and characteristic expressions associated with the current U.S. President.

✨ Model Details

Feature	Description
Base Model	mistralai/Mistral-7B-Instruct-v0.2
Architecture	LoRA adapter (Low-Rank Adaptation)
LoRA Rank	16
Language	English
Training Focus	Biden's communication style, rhetoric, and response patterns
Merged Adapters	Combines style and identity LoRA weights from: - nnat03/biden-mistral-adapter (original adapter) - ./identity-adapters/biden-identity-adapter

🎯 Intended Use

📚 Education	🔍 Research	🎭 Creative
Political discourse analysis	Rhetoric pattern studies	Interactive simulations

📊 Training Data

This model was trained on carefully curated datasets that capture authentic speech patterns:

📱 Biden tweets dataset (2007-2020) - Extensive collection capturing everyday communication
🎤 Biden 2020 DNC speech dataset - Formal oratorical patterns

These datasets were processed into a specialized instruction format to optimize learning of distinctive speech patterns.

⚙️ Technical Specifications

Training Configuration

🧠 Framework: Hugging Face Transformers + PEFT
📊 Optimization: 4-bit quantization
🔧 LoRA Config: r=16, alpha=64, dropout=0.05
🎛️ Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

Training Parameters

📦 Batch size: 4
🔄 Gradient accumulation: 4
📈 Learning rate: 2e-4
🔁 Epochs: 3
📉 LR scheduler: cosine
⚡ Optimizer: paged_adamw_8bit
🧮 Precision: BF16

⚠️ Limitations and Biases

This model mimics a speaking style but doesn't guarantee factual accuracy
While emulating Biden's rhetoric, it doesn't represent his actual views
May reproduce biases present in the training data
Not suitable for production applications without additional fact-checking

💻 Usage

Run this code to start using the adapter with the Mistral-7B-Instruct-v0.2 base model:

python

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel
import torch

# Load base model with 4-bit quantization
base_model_id = "mistralai/Mistral-7B-Instruct-v0.2"
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
)

# Load model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    quantization_config=bnb_config,
    device_map="auto", 
    torch_dtype=torch.float16
)
tokenizer = AutoTokenizer.from_pretrained(base_model_id)

# Apply the adapter
model = PeftModel.from_pretrained(model, "nnat03/biden-mistral-adapter")

# Generate a response
prompt = "What's your vision for America's future?"
input_text = f"<s>[INST] {prompt} [/INST]"
inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_length=512, temperature=0.7, do_sample=True)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response.split("[/INST]")[-1].strip())

📚 Citation

If you use this model in your research, please cite:

bibtex

@misc{nnat03-biden-mistral-adapter,
  author = {nnat03},
  title = {Biden Mistral Adapter},
  year = {2023},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/nnat03/biden-mistral-adapter}}
}

🔍 Ethical Considerations

This model is created for educational and research purposes. It attempts to mimic the speaking style of a public figure but does not represent their actual views or statements. Use responsibly.

Framework version: PEFT 0.15.0

Made with ❤️ for NLP research and education

SecretLab/biden-mistral-adapter

🇺🇸 Biden Mistral Adapter 🇺🇸