Base model: microsoft/Phi-3-mini-4k-instruct
Method: QLoRA (4-bit NF4) + SFT → DPO
SFT data: Single-turn and multi-turn Q&A grounded in canonical texts
DPO data: Preference pairs (verified teachings vs. generic responses)
Canonical sources: Who Am I?, Talks with Sri Ramana Maharshi, Ulladu Narpadu
LoRA rank: 16 | Alpha: 32 | Target: all attention + MLP projection layers

Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id  = "SriRamanaAtmic/AtmicIntelv1"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model     = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype       = torch.bfloat16,
    device_map        = "auto",
    trust_remote_code = True,
)

SYSTEM = (
    "You are a knowledgeable and compassionate guide to the teachings of "
    "Sri Ramana Maharshi. Answer questions about Self-enquiry, the nature "
    "of the Self, surrender, and the path to liberation, grounded in his "
    "actual teachings."
)

messages = [
    {"role": "system",    "content": SYSTEM},
    {"role": "user",      "content": "What is self-enquiry?"},
]
text   = tokenizer.apply_chat_template(messages, tokenize=False,
                                        add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.no_grad():
    out = model.generate(
        **inputs,
        max_new_tokens     = 512,
        temperature        = 0.7,
        top_p              = 0.9,
        repetition_penalty = 1.1,
        do_sample          = True,
    )
print(tokenizer.decode(out[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

About Sri Ramana Maharshi

Sri Ramana Maharshi (1879–1950) was one of the greatest sages of modern India. His principal teaching was Atma Vichara (Self-enquiry): the direct path of tracing the sense of "I" back to its source, the Self — pure, undivided awareness. He is revered across traditions for the simplicity, depth, and transformative power of his teachings.