DeepSeek Coder 1.3B Tutorial

---

1. Introduction

DeepSeek Coder is a powerful open-source code model designed for project-level code generation and infilling. It supports multiple programming languages and achieves state-of-the-art results in code completion tasks.

2. Features of DeepSeek Coder

• Trained on 2T tokens (87% code, 13% natural language in English & Chinese).
• Available in multiple sizes (1.3B, 5.7B, 6.7B, and 33B parameters).
• 16K context window for handling large projects.
• Supports project-level code completion & infilling.

---

3. How to Use DeepSeek Coder 1.3B

A. Installing Required Dependencies

To use the model in Python, install the necessary libraries:

pip install transformers torch

B. Loading the Model in Python

Use the transformers library to load the model:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/deepseek-coder-1.3b-instruct", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    "deepseek-ai/deepseek-coder-1.3b-instruct",
    trust_remote_code=True,
    torch_dtype=torch.bfloat16
).cuda()  # Use GPU for faster inference

C. Running Code Generation

You can prompt the model to generate code. Here’s an example using QuickSort:

messages = [
    {"role": "user", "content": "Write a quick sort algorithm in Python."}
]

inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
outputs = model.generate(
    inputs,
    max_new_tokens=512,
    do_sample=False,
    top_k=50,
    top_p=0.95,
    num_return_sequences=1,
    eos_token_id=tokenizer.eos_token_id
)

print(tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True))

This will generate a Python implementation of QuickSort.

---

4. Running DeepSeek Coder Locally

If you want to run the model efficiently on your system, consider using GGUF format models with llama.cpp or text-generation-webui.

A. Running with llama.cpp

./main -m deepseek-coder-1.3b-instruct.gguf -p "Write a Python function to check if a number is prime."

B. Running with text-generation-webui

1. Download the GGUF model from DeepSeek Repository.
2. Load it into text-generation-webui and start inference.

---

5. License and Commercial Use

• The code is licensed under MIT.
• The model has a separate Model License allowing commercial use.

---

6. Conclusion

DeepSeek Coder 1.3B is a powerful AI coding assistant that supports code generation, completion, and infilling. It can be run on your local machine with PyTorch or optimized with GGUF format for efficiency.