📌 Step 1: Google Cloud Setup

1.1 Sign Up for Google Cloud & Enable Billing

1. Go to Google Cloud Console and create an account.
2. Set Up Billing:
   • Navigate to Billing → Create Billing Account.
   • Attach it to your new project.
   • Google provides free credits for new users.

1.2 Create a New Project

1. In Google Cloud Console, click Select a Project → New Project.
2. Name it (e.g., deepseek-ai).
3. Click Create.

1.3 Enable Compute Engine API

1. Navigate to APIs & Services → Enable APIs.
2. Search for Compute Engine API and enable it.

📌 Step 2: Create Google Cloud VM with T4 GPU

2.1 Create a Virtual Machine Instance

1. Go to Compute Engine → VM Instances → Create Instance.
2. Configure instance settings:
   • Name: deepseek-t4
   • Region: us-central1
   • Machine Type: n1-standard-8 (8 vCPUs, 30GB RAM)
   • GPU: 1 x NVIDIA T4
   • Boot Disk: Ubuntu 22.04 LTS, 100GB SSD
3. Click Create and wait for the VM to start.

2.2 Connect to the VM via SSH

gcloud compute ssh deepseek-t4

📌 Step 3: Install Required Dependencies

3.1 Update the System & Install CUDA

sudo apt update && sudo apt upgrade -y
sudo apt install -y nvidia-cuda-toolkit
nvidia-smi
            

3.2 Install Python & Virtual Environment

sudo apt install -y python3 python3-pip python3-venv
python3 -m venv deepseek-env
source deepseek-env/bin/activate
            

3.3 Install Required Python Packages

pip install torch torchvision transformers accelerate flask bitsandbytes
            

📌 Step 4: Load DeepSeek 14B

4.1 Full Python Code for Running the Model

import os
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B"
huggingface_token = os.getenv("HUGGINGFACE_TOKEN")

tokenizer = AutoTokenizer.from_pretrained(model_name, token=huggingface_token)
model = AutoModelForCausalLM.from_pretrained(
    model_name, token=huggingface_token, device_map="auto", torch_dtype=torch.float16
)

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

outputs = model.generate(
    inputs.input_ids.to(model.device),
    attention_mask=inputs.attention_mask.to(model.device),
    max_length=200,
    pad_token_id=tokenizer.eos_token_id
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print("\nResponse:", response)