🚀 EduMentor-Qwen3-4B-FP16

EduMentor is a fine-tuned Qwen3 4B based AI engineering mentor designed for real-time conversational learning, coding assistance, project guidance, and career mentoring.

Unlike a normal chatbot, EduMentor is optimized for speech-to-speech AI systems, where the model separates spoken responses from visual artifacts like code, diagrams, workflows, and roadmaps.

The goal is simple:

An AI mentor that talks like a human teacher while still producing structured engineering resources.

---

✨ Key Features

🎙 Speech-to-Speech Optimized Responses

EduMentor follows a structured response format:

{
  "speech": "Short natural explanation spoken through TTS",
  "display": {
    "type": "code | diagram | roadmap | notes",
    "content": "Detailed visual artifact"
  },
  "follow_up": "Context-aware continuation question"
}

This allows integration with:

Voice assistants
Real-time AI tutors
Multimodal agents
Learning platforms
🧠 Engineering Mentor Capabilities

EduMentor focuses on engineering education across multiple domains:

💻 Computer Science
Programming concepts
Data structures and algorithms
Debugging help
System design
AI/ML concepts
Software engineering
🤖 Artificial Intelligence
Machine learning
Deep learning
LLM concepts
Model deployment
RAG pipelines
⚡ Electronics / ECE
Digital electronics
MOSFETs
Circuits
Embedded basics
⚙ Mechanical Engineering
Engines
Manufacturing processes
Core engineering concepts
🏗 Civil Engineering
RCC concepts
Structural basics
Engineering fundamentals
🎯 Career Mentor Features

EduMentor can assist students with:

Placement preparation
Internship planning
Resume improvement
Project ideas
Learning roadmaps
Interview preparation
Skill development
🏗 Model Architecture

Base:

Qwen3-4B-Instruct-2507

Fine-tuning:

Method: LoRA SFT
Rank: 32
Domain: Engineering mentoring + Voice interaction

Training focused on:

Natural mentor conversations
Engineering explanations
Code artifact separation
Emotional support
Career guidance
Multi-domain engineering knowledge
📦 Available Versions
FP16

This repository contains the full merged FP16 checkpoint.

Recommended for:

Further fine-tuning
vLLM deployment
High quality inference
GGUF Quantized Versions

Optimized GGUF builds are available separately:

Q6_K → Higher quality inference
Q4_K_M → Local real-time voice deployment

Recommended stack:

EduMentor-Qwen3-GGUF
        +
llama.cpp
        +
GLM-4-Voice
🚀 Usage

Install dependencies:

pip install transformers accelerate torch

Load model:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch


model_id = "PraneetNS/EduMentor-Qwen3-4B-FP16"


tokenizer = AutoTokenizer.from_pretrained(
    model_id
)


model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)


messages = [
    {
        "role":"system",
        "content":"You are EduMentor, an AI engineering mentor."
    },
    {
        "role":"user",
        "content":"Explain recursion simply"
    }
]


prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)


inputs = tokenizer(
    prompt,
    return_tensors="pt"
).to(model.device)


output=model.generate(
    **inputs,
    max_new_tokens=300
)


print(
tokenizer.decode(output[0])
)
🧪 Example Output

User:

Write binary search code in Python

EduMentor:

{
"speech":
"I created the implementation below. Let's understand the idea first.",

"display":
{
"type":"code",
"language":"python",
"content":"def binary_search(...)"
},

"follow_up":
"Would you like to understand the time complexity?"
}
🔊 Intended Voice Pipeline

EduMentor was designed for:

User Speech

↓

Speech Encoder

↓

EduMentor Qwen3

↓

JSON Router

↙              ↘

TTS Speech     Display Renderer

This prevents:

❌ Reading code aloud
❌ Speaking large tables
❌ Narrating diagrams

while preserving natural conversations.

⚠️ Limitations
May require a strong system prompt for strict artifact routing.
Engineering answers should be verified for critical applications.
Not intended to replace certified professional advice.
🌱 Future Roadmap

Planned improvements:

Larger engineering datasets
Tool calling
RAG integration
Real-time project assistant
Personalized student memory
Multimodal tutoring
👨‍💻 Creator

Built as an experiment toward creating a personalized AI mentor for engineering students.

EduMentor aims to make high-quality technical guidance accessible through natural AI conversations.


Then commit it:

```python
from huggingface_hub import upload_file


upload_file(
    path_or_fileobj="README.md",
    path_in_repo="README.md",
    repo_id="PraneetNS/EduMentor-Qwen3-4B-FP16",
    repo_type="model",
    commit_message="Add EduMentor model card"
)