GREEN|CSV[.1]

>_Sandbox >_About

Tutorial: Setting up Mellum focal model on VsCode with Continue

tutorial ollama continue vs-Code jetbrains

With Mellum launching as an open source model1, I found a good opportunity to test it against my current Unity learning2 workflow. Rider/Unity runs slowly, so VS Code is my go to, and this week I have read two posts that align with that flow. Resharper is in public preview in VS Code 3, and Mellum is now open source1.

Tutorial: Setting Up and Running Mellum

  1. Clone the Mellum model files
   git clone https://huggingface.co/JetBrains/Mellum-4b-base-gguf

(~4.5 GB)1

  1. Install Ollama4 Download and install the appropriate Ollama package for your OS (Ollama, 2025).

  2. Install Continue5 in VS Code Add the Continue extension to wire up Mellum via Ollama.

Add Mellum to Ollama

  1. In your project directory, create a file named Modelfile with:
FROM ./mellum-4b-base.Q8_0.gguf
# PARAMETER <parameter> <value>

SYSTEM "You are a completion assistant. Predict the most likely continuation of the given input accurately and concisely."

PARAMETER temperature      0.2
PARAMETER top_p            0.9
PARAMETER top_k            40
PARAMETER repeat_penalty   1.2
PARAMETER repeat_last_n    128
PARAMETER num_predict      512
PARAMETER num_ctx          8192

PARAMETER stop "###"
PARAMETER stop "\n\n"

All of these directives and parameters come from the Ollama Modelfile6

  1. (Optional) To run a quantized build on modest hardware at the expense of some accuracy, see Ollama’s quantization guide:

https://ollama.readthedocs.io/en/import/#quantizing-a-model

  1. Create your Ollama model
ollama create Mellum -f .\Modelfile

Make sure the FROM base matches the one used to the Model file (Chang, n.d.).

  1. Inspect default values
ollama show Mellum

You’ll see something like:

 Model
    architecture        llama
    parameters          4.0B
    context length      8192
    embedding length    3072
    quantization        Q8_0

  Capabilities
    completion

  Parameters
    top_k             40
    top_p             0.9
    num_ctx           8192
    num_predict       512
    repeat_last_n     128
    repeat_penalty    1.2
    stop              "###"
    stop              "\n\n"
    temperature       0.2

Tweak any settings in your Modelfile, then rebuild with ollama create as above.

VS Code, Continue & Ollama

Continue5 lets you hook any local or remote model into VS Code, select Ollama as your provider, then Add a Local Model even though it isn’t in the Ollama registry by default. Configure per-context assistants from the Continue UI, and you’ll get Mellum focal context completions right in VS Code.

Continue configuration for Ollama + Mellum

Overall, on my station with a dedicated GPU it’s fast, but on smaller machines you can quantize the model to fit your needs. Now I can zip between Unity and VS Code with context-aware completions via Mellum.

In continue configured for each context available here are some insights

Your browser doesn’t support video.

Your browser doesn’t support video.

67189104235

Footnotes

  1. JetBrains. (2025b, April). Mellum goes open source: A purpose‐built LLM for developers, now on Hugging Face. JetBrains Blog. Retrieved May 21, 2025, from https://blog.jetbrains.com/ai/2025/04/mellum-goes-open-source-a-purpose-built-llm-for-developers-now-on-hugging-face/ ↩ ↩2 ↩3 ↩4

  2. YouTube. (n.d.). How To Make a 3D Space Shooter Game in Unity – Tutorial [Video]. YouTube. Retrieved May 21, 2025, from https://www.youtube.com/watch?v=VW3PkEF1Fzk ↩ ↩2

  3. JetBrains. (2025f, May). ReSharper for Visual Studio Code [Extension]. Visual Studio Marketplace. Retrieved May 21, 2025, from https://marketplace.visualstudio.com/items?itemName=jetbrains.resharper-code ↩ ↩2

  4. Ollama. (2025). Download Ollama on Windows [Software]. Retrieved May 21, 2025, from https://ollama.com/download/windows ↩ ↩2

  5. Continue.dev. (2025). Continue – Open-source AI code assistant [Extension]. Visual Studio Marketplace. Retrieved May 21, 2025, from https://marketplace.visualstudio.com/items?itemName=Continue.continue ↩ ↩2 ↩3

  6. Chang, L. (n.d.). Model file specification Ollama [Documentation]. GitHub. Retrieved May 21, 2025, from https://github.com/lloydchang/ollama-ollama/blob/main/docs/modelfile.md ↩ ↩2

  7. JetBrains. (2025a, February). Why and how JetBrains built Mellum – the LLM designed for code completion. JetBrains Blog. Retrieved May 21, 2025, from https://blog.jetbrains.com/ai/2025/02/why-and-how-jetbrains-built-mellum-the-llm-designed-for-code-completion/ ↩

  8. JetBrains. (2025c, April 8). JetBrains Mellum overview & usage [IDE Services documentation]. Retrieved May 21, 2025, from https://www.jetbrains.com/help/ide-services/jetbrains-mellum.html ↩

  9. JetBrains. (2025d, May). JetBrains AI Assistant – now in Visual Studio Code. JetBrains Blog. Retrieved May 21, 2025, from https://blog.jetbrains.com/ai/2025/05/jetbrains-ai-assistant-now-in-visual-studio-code/ ↩

  10. JetBrains. (2025e). Mellum-4b-base-gguf [Model]. Hugging Face. Retrieved May 21, 2025, from https://huggingface.co/JetBrains/Mellum-4b-base-gguf ↩