EvoNet/EvoNet-1.5B-v0.1-Alpha

🚀 EvoNet-1.5B v0.1 Alpha (Lite Series)

"Trí tuệ của mô hình 3B trong hình hài nhỏ gọn." Developed by EvoNet Technology - Founder: Huỳnh Dương Phong.

EvoNet-1.5B is a specialized Vietnamese-English Code LLM. This is the first Alpha release in the EvoNet Ecosystem, designed to bring high-level reasoning to low-resource devices.

💡 Concept & Architecture (Why 1.5B?)

Although built on the robust architecture of Qwen2.5-3B, this model is part of our "Lite Series" experiments.

We designate it as EvoNet-1.5B to reflect its target deployment environment and resource footprint optimization. Through advanced quantization and optimization techniques (GGUF/4-bit), we aim to deliver 3B-level coding intelligence that runs as efficiently as a traditional 1.5B model on consumer CPUs and Edge Devices.

🌟 Key Features

• 🇻🇳 Vietnamese-Centric Coding: Finely tuned to understand and explain complex Python logic in natural Vietnamese, specifically catering to Vietnamese developers.
• 🧠 High-Density Intelligence: Leverages the deep reasoning capabilities of a 3B base model, outperforming standard 1.5B models in logic and instruction following.
• ⚡ CPU Optimized: Engineered to run smoothly on free-tier Spaces (CPU only) and personal laptops without requiring heavy GPUs.
• 🆔 Identity Awareness: Fully self-aware of its origin (EvoNet Technology) and creator.

🛠️ Technical Specifications

• Base Architecture: Qwen/Qwen2.5-3B-Instruct
• Target Size Class: ~1.5B (via Quantization/Efficiency optimization)
• Dataset: EvoNet-Corpus-50k (A curated mix of Synthetic Logic, Real-world Python Code, and Vietnamese Instructions).
• Training Method: LoRA Fine-tuning via Unsloth (Accelerated & Efficient).

🚀 Roadmap & Grants

This Alpha release serves as a proof-of-concept for the EvoNet architecture.

• Current: Optimized CPU Inference (v0.1 Alpha).
• Future Goal: We are actively applying for GPU Grants to scale this architecture to a full EvoNet-7B version, aiming for state-of-the-art performance in the Vietnamese coding domain.

⚠️ Limitations (Alpha v0.1)

• Hallucination: As an Alpha release, it may occasionally generate incorrect logic for highly complex queries.
• Scope: Primarily optimized for Python and General Chat; other languages may not be as fluent.
• Usage: Intended for research, testing, and educational purposes.

---

© 2026 EvoNet Technology. Open Source for the Community.