Roy-v1

Roy is a personal AI assistant model created and fine-tuned by Souvik Pramanick.
Designed to be helpful, conversational, and practical for everyday tasks such as learning, coding, problem solving, and general assistance.

---

Creator

Founder & Trainer:
Souvik Pramanick
GitHub: https://github.com/Souvik18p
HuggingFace: https://huggingface.co/souvik18

Roy is an independent project built with the vision of creating a smart, friendly, and customizable AI assistant.

---

What Roy Can Do

Roy is capable of:

• Natural conversation and assistance
• Answering general knowledge questions
• Solving math and logical problems
• Helping with coding and debugging
• Writing emails, stories, and content
• Explaining concepts in simple language
• Brainstorming ideas and learning support

---

Model Details

• Model Name: Roy-v1
• Parameters: 7B
• Architecture: LLaMA-based
• Tensor Type: F16
• Format: Safetensors
• License: Open for community usage

Base Model: souvik18/Roy-v1

Quantized Versions (Community)

Thanks to @mradermacher for providing GGUF quants of Roy-v1:

https://huggingface.co/mradermacher/Roy-v1-GGUF

These versions allow Roy to run on:

• CPU only systems
• Low VRAM GPUs
• Mobile / local apps via llama.cpp, ollama, koboldcpp

Quick Usage

Using HuggingFace Transformers

!pip install -U transformers datasets accelerate bitsandbytes peft huggingface_hub

from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import torch

MODEL_ID = "souvik18/Roy-v1"

# 4bit config – works best on Kaggle
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_use_double_quant=True,
)

print(" Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
tokenizer.pad_token = tokenizer.eos_token

print(" Loading model (4bit)...")
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    quantization_config=bnb_config,
    device_map="auto"
)

print("\n Roy-v1 Loaded Successfully!")

while True:
    text = input("You: ")
    if text.lower() in ["exit","quit"]:
        break

    prompt = f"[INST] {text} [/INST]"

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

    with torch.no_grad():
        out = model.generate(
            **inputs,
            max_new_tokens=200,
            temperature=0.7,
            top_p=0.9,
            do_sample=True
        )