Property	Value
Base Model	Qwen/Qwen2.5-3B-Instruct
Method	LoRA (Low-Rank Adaptation)
Rank	16
Alpha	32
Target Modules	q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Training Precision	bf16
Quantization	4-bit (BitsAndBytes NF4)
Trainable Params	~29M / 3.1B (0.96%)

Training Details

Property	Value
Instance	AWS SageMaker ml.g5.2xlarge (NVIDIA A10G 22 GB)
Epochs	5
Batch Size	2 (gradient accumulation 8, effective 16)
Learning Rate	2e-4 (cosine schedule)
Optimizer	AdamW 8-bit
Best Epoch	3 (val loss 0.1838)

Training Loss

Epoch	Training Loss	Validation Loss
1	0.3478	0.2551
2	0.1432	0.1891
3	0.1020	0.1838
4	0.0835	0.1930

Usage

Load with PEFT (recommended)

python

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

base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2.5-3B-Instruct",
    torch_dtype=torch.float16,
    device_map="auto",
)
model = PeftModel.from_pretrained(base_model, "bhaiyahnsingh45/bhaiya-loan-assistant-lora")
model.eval()

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-3B-Instruct")

messages = [
    {"role": "system", "content": "You are a compliant and helpful loan assistant for Bhaiya & Company — Banking & Finance Division."},
    {"role": "user", "content": "My salary is 55,000, CIBIL 740, age 30. Am I eligible for a personal loan?"},
]

inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(
        input_ids=inputs["input_ids"],
        max_new_tokens=512,
        temperature=0.3,
        do_sample=True
    )

generated_tokens = outputs[0][inputs["input_ids"].shape[1]:]

print(tokenizer.decode(generated_tokens, skip_special_tokens=True))

Intended Use

This model is a loan assistant for Bhaiya & Company — Banking & Finance Division. It handles:

Loan eligibility checks (Personal, Business, Home loans)
Document requirement guidance
EMI calculations
Application process guidance

Limitations

Never guarantees loan approval
Refuses investment advice, stock tips, financial planning
Final approval is always subject to verification

Dataset

Trained on bhaiyahnsingh45/bhaiya-loan-assistant-dataset — a custom instruction-tuning dataset in chat format (system/user/assistant messages).

bhaiyasingh45/bhaiya-loan-assistant-lora

Bhaiya & Company — Loan Assistant (LoRA Adapter)

Model Details