Instructions to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF", dtype="auto")

llama-cpp-python

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF",
	filename="Devstral-Small-2-24B-Instruct-2512-IQ2_S-2.34bpw.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S
# Run inference directly in the terminal:
llama-cli -hf byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S
# Run inference directly in the terminal:
llama-cli -hf byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

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 byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S
# Run inference directly in the terminal:
./llama-cli -hf byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

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 byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

Use Docker

docker model run hf.co/byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

LM Studio
Jan

vLLM

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

SGLang

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with Ollama:
```
ollama run hf.co/byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S
```

Unsloth Studio new

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF to start chatting

Pi new

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

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 byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

Run Hermes

hermes

Docker Model Runner
How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with Docker Model Runner:
```
docker model run hf.co/byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S
```

Lemonade

How to use byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull byteshape/Devstral-Small-2-24B-Instruct-2512-GGUF:IQ2_S

Run and chat with the model

lemonade run user.Devstral-Small-2-24B-Instruct-2512-GGUF-IQ2_S

List all available models

lemonade list

Exceptional Stability and Performance Quantized Devstral Small 2 (IQ4_XS) at 10K Context

by Kamil21322 - opened Mar 2

Discussion

Kamil21322

Mar 2

I would like to share my experience and sincere appreciation for the quantized Devstral Small 2 (IQ4_XS) model.

The performance has been outstanding. At a 10K context length, the model consistently runs at approximately 30 tokens per second, which is highly impressive for a quantized configuration. More importantly, it maintains coherence and logical consistency as the context fills. It does not degrade into irrelevant or nonsensical output, which is often a concern with extended contexts.

I also experimented with setting the KV cache to Q5.1 and increasing the context length further. Even under these conditions, the model preserved its stability. I conducted multiple tests across different scenarios (which I won’t be sharing here as they are part of my own projects), and the results were consistently strong. The reliability and balance between efficiency and quality are genuinely remarkable.

When the context limit is fully reached, the model naturally stops upon receiving a new prompt, which is expected behavior. Up until that limit, however, it performs flawlessly.

Your quantized models clearly reflect high-level optimization and engineering excellence. The balance between speed, memory efficiency, and output quality is extremely well executed.

I am also eagerly looking forward to testing the newly released Qwen 3.5 27B and 35B models. If they follow the same level of optimization and stability, they will be absolutely impressive.

My sincere congratulations to the entire team — your work is truly commendable.

BingoBird

Mar 27

•

edited Mar 27

Try Qwen3-Coder-30B-A3B-Instruct-Q3_K_S-2.69bpw.gguf too! <3

Kamil21322

Apr 4

•

edited Apr 4

Byteshape quantization works great; they do BF16-quality quantization, but the models themselves don't meet my needs. :(

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