Instructions to use unsloth/DeepSeek-V3.2-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use unsloth/DeepSeek-V3.2-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./llama-cli -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

LM Studio
Jan
Ollama
How to use unsloth/DeepSeek-V3.2-GGUF with Ollama:
```
ollama run hf.co/unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL
```
Unsloth Desktop

How to use unsloth/DeepSeek-V3.2-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use unsloth/DeepSeek-V3.2-GGUF with Docker Model Runner:
```
docker model run hf.co/unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL
```

Lemonade

How to use unsloth/DeepSeek-V3.2-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.DeepSeek-V3.2-GGUF-UD-Q4_K_XL

List all available models

lemonade list

Hermes Agent

How to use unsloth/DeepSeek-V3.2-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use unsloth/DeepSeek-V3.2-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Issues with reasoning content and Open WebUI

by zettabomb - opened Feb 6

Discussion

zettabomb

Feb 6

I mentioned this on the Reddit announcement post but I've tested it a bit more and figured here was a better place for it.

I was previously using the UD-Q4_K_XL quant of DeepSeek-V3.1-Terminus with ik_llama.cpp and getting good results, with Open WebUI as the frontend. Switching to DeepSeek-V3.2 (still at UD-Q4_K_XL) causes issues with the reasoning content, however. One way or another, Open WebUI doesn't "see" the starting tag and just puts the reasoning content in as regular text, including the closing . Using the V3.1-Terminus chat template with V3.2 seems to work, but then breaks tool calling support. Upon investigation, ik_llama.cpp reports that the actual prompt_tokens are identical - for my standard test prompt I see exactly "<｜begin▁of▁sentence｜><｜User｜>What is the airspeed velocity of an unladen swallow?<｜Assistant｜>" regardless of chat template.

I don't immediately see a reason for the discrepancy. It seems as if potentially it's an issue with ik_llama.cpp's internal parsing of the output into JSON, but it could equally be with Open WebUI's parsing, or with the chat template. Anything I can try to fix this?

tnuvkeg

Feb 9

With ud-q3k I also see that the thinking process is bleeding into the answer:

"Hmm, the user just said "Hello," which is a simple greeting. This doesn't require any complex analysis or deep thought.
I should respond warmly and professionally to set a positive tone for the conversation. Since this is an initial greeting, it's best to keep it friendly and open-ended to encourage further interaction.
A straightforward "Hello!" with an offer of assistance seems appropriate here. No need to overcomplicate it.Hello! How can I assist you today?"

This (including the same text but in Chinese, even when the same "Hello" prompt) happens to me with both Open Webui and llama.cpp's own UI and with llama.cpp and ik_llama.cpp.

"Hiding" the thinking process works fine in other models (kimi-k2.5, glm-4.7-flash, etc)

About the chat template from Terminus, doesn't that disable thinking? because I've just tried it and while I don't see the thinking process, I also don't see a way to show it, so I think "thinking" is disabled by it.

haoshenghan

Feb 9

•

This comment has been hidden (marked as Resolved)

tnuvkeg

Feb 10

Hi haoshenghan,
yes, I always use "--jinja" in all Unsloth's quants.

About the chat template, that's how I tried and it just disabled reasoning. So yes, I don't see I thinking block... because there is no thinking :D

What inference engine and UI, and quant are you using? Because I tried with llama.cpp's own UI and I have the same issue (also with ik_llama.cpp), but only with this model.

haoshenghan

Feb 10

•

This comment has been hidden (marked as Resolved)

zettabomb

Feb 12

So, I have already tried using the V3.1 template. It does seem to work ok, and I edited it to turn on thinking by default (although I think that can also be done with a CLI argument). I don't think it works correctly with tools though. Did they change their tool calling format?

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment