Qwen2.5-1.5B-Financial-Code-Engine
An end-to-end post-trained Large Language Model optimized for domain-specific financial reasoning, JSON formatting, and bug-free Python code execution logic.
This model was adapted from Qwen/Qwen2.5-1.5B-Instruct using LLaMA-Factory through a 2-stage post-training pipeline:
- Phase 1 — Supervised Fine-Tuning (SFT): 4-bit QLoRA training on financial Q&A and Python code datasets.
- Phase 2 — Direct Preference Optimization (DPO): Preference alignment using pairwise preference data to guide model outputs toward concise, well-commented code and accurate financial reasoning.
Model Details
- Developed by: Ahmad Nawaz (LinkedIn Profile)
- Model Type: Causal Language Model (Fine-tuned & DPO Aligned)
- Base Model:
Qwen/Qwen2.5-1.5B-Instruct - Language(s): English, Python
- License: Apache-2.0
- Fine-Tuning Framework: LLaMA-Factory (QLoRA)
- Alignment Method: Direct Preference Optimization (DPO)
Uses
Direct Use
This model is designed for:
- Financial reasoning and financial calculation logic (e.g., NPV, CAGR, loan schedules).
- Python code generation, syntax debugging, and structured commentary.
- Structured JSON and entity extraction for downstream AI agents or API pipelines.
Out-of-Scope Use
This model should not be relied upon as certified legal or financial investment advice.
How to Get Started with the Model
You can run inference using Hugging Face transformers:
import torch
from transformers import pipeline
# Load pipeline directly from Hugging Face Hub
pipe = pipeline(
"text-generation",
model="ahmadnawaz21/Qwen2.5-1.5B-Financial-Code-Engine",
torch_dtype=torch.float16,
device_map="auto"
)
# ChatML prompt structure
prompt = """<|im_start|>system
You are an expert financial and software engineering assistant.<|im_end|>
<|im_start|>user
Write a Python function to calculate the Net Present Value (NPV) given a discount rate and a list of cash flows.<|im_end|>
<|im_start|>assistant
"""
outputs = pipe(
prompt,
max_new_tokens=256,
do_sample=True,
temperature=0.7,
top_p=0.9
)
print(outputs[0]["generated_text"])