# DeepSeek-R1-Distill-Llama-8B Fine-tuned on Finance Dataset

This model is a fine-tuned version of DeepSeek-R1-Distill-Llama-8B using LoRA adapters, trained on financial instruction data.

## Model Details
- **Base Model**: unsloth/DeepSeek-R1-Distill-Llama-8B-unsloth-bnb-4bit
- **Fine-tuning Method**: LoRA
- **LoRA Parameters**: r=4, alpha=16
- **Target Modules**: q_proj, k_proj, v_proj, o_proj
- **Training Dataset**: Rishi-19/finance-instruct-dataset

## Use Cases
This model is optimized for financial analysis, valuation calculations, and financial advisory tasks.

## Example Usage
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel, PeftConfig
import torch

# Set device
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load the model and tokenizer
model_name = "Rishi-19/deepseek_finetuned_model_rishi"
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Load the base model first
peft_config = PeftConfig.from_pretrained(model_name)
base_model = AutoModelForCausalLM.from_pretrained(
    peft_config.base_model_name_or_path, 
    torch_dtype=torch.float16,  # Use half precision to save memory
    device_map="auto",
    trust_remote_code=True
)

# Then load the PEFT adapter
model = PeftModel.from_pretrained(base_model, model_name)
model.eval()  # Set to evaluation mode

# Generate text
inputs = tokenizer("Calculate the Net Present Value of a project with initial investment of $1M", return_tensors="pt").to(device)
with torch.no_grad():
    outputs = model.generate(**inputs, max_length=200)
print(tokenizer.decode(outputs[0]))
```