murli-assistant-phi2-lora

Murli Assistant - Fine-tuned Phi-2 with LoRA

This model is a fine-tuned version of microsoft/phi-2 using LoRA (Low-Rank Adaptation) on Brahma Kumaris Murli data.

Model Description

• Base Model: microsoft/phi-2 (2.7B parameters)
• Fine-tuning Method: LoRA (r=8, alpha=16)
• Training Data: 100+ daily murlis from MongoDB database
• Use Case: Spiritual guidance and murli knowledge assistant

Training Details

• LoRA Rank (r): 8
• LoRA Alpha: 16
• Target Modules: q_proj, o_proj, k_proj, v_proj
• Training Examples: 201 formatted instructions
• Adapter Size: ~15MB

Usage

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

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "microsoft/phi-2",
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "eswarankrishnamurthy/murli-assistant-phi2-lora")
tokenizer = AutoTokenizer.from_pretrained("microsoft/phi-2")
tokenizer.pad_token = tokenizer.eos_token

# Generate response
question = "What is the essence of today's murli?"
prompt = f"### Instruction:\n{question}\n\n### Response:\n"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=200)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Inference API

This model is also available via Hugging Face Inference API:

import requests

API_URL = "https://api-inference.huggingface.co/models/eswarankrishnamurthy/murli-assistant-phi2-lora"
headers = {"Authorization": f"Bearer {YOUR_HF_TOKEN}"}

def query(payload):
    response = requests.post(API_URL, headers=headers, json=payload)
    return response.json()

output = query({"inputs": "What is soul consciousness?"})
print(output)

Training Information

The model was trained on diverse murli content including:

• Daily murli essence
• Blessings and slogans
• Questions and answers
• Spiritual teachings and guidance

Limitations

• Best performance on spiritual/murli-related queries
• May require GPU for faster inference
• CPU inference is possible but slower

Citation

If you use this model, please cite:

@misc{murli-assistant-phi2,
  author = {eswarankrishnamurthy},
  title = {Murli Assistant - Fine-tuned Phi-2},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/eswarankrishnamurthy/murli-assistant-phi2-lora}
}

Contact

For questions or feedback, please open an issue on the model repository.