---

license: apache-2.0

base_model: mistralai/Shieldstral-1.0-3B

tags:

- gguf

- llama.cpp

- shieldstral

- mistral

- quantized

pipeline_tag: text-generation

language:

- en

---

Shieldstral-1.0-3B-GGUF

GGUF conversions and quantizations of mistralai/Shieldstral-1.0-3B for use with:

• llama.cpp
• LM Studio
• Ollama
• KoboldCpp
• Jan
• Text Generation WebUI
• Other GGUF-compatible runtimes

---

Available Files

Full Precision

• Shieldstral-1.0-3B-F16.gguf

Quantized Versions

• Shieldstral-1.0-3B-Q4_K_M.gguf
• Shieldstral-1.0-3B-Q5_K_M.gguf
• Shieldstral-1.0-3B-Q6_K.gguf
• Shieldstral-1.0-3B-Q8_0.gguf

---

Recommended Quant

For most users:

Shieldstral-1.0-3B-Q4_K_M.gguf

Provides the best balance between:

• Quality
• Memory usage
• Inference speed

---

File Sizes

Approximate sizes:

F16      ~6.9 GB
Q4_K_M   ~2.2 GB
Q5_K_M   ~2.5 GB
Q6_K     ~2.8 GB
Q8_0     ~3.5 GB

---

llama.cpp

Example:

./llama-cli \
  -m Shieldstral-1.0-3B-Q4_K_M.gguf \
  -p "Hello"

Chat mode:

./llama-cli \
  -m Shieldstral-1.0-3B-Q4_K_M.gguf \
  -cnv

---

LM Studio

1. Download your preferred GGUF file.
2. Open LM Studio.
3. Import the model.
4. Start chatting.

---

Ollama

Create a file named Modelfile:

FROM Shieldstral-1.0-3B-Q4_K_M.gguf

Then run:

ollama create shieldstral -f Modelfile
ollama run shieldstral

---

Base Model

Base model:

mistralai/Shieldstral-1.0-3B

All model architecture, training, licensing, benchmarks, and usage rights belong to Mistral AI.

Please refer to the original model repository for official documentation.

---

Conversion Notes

The model was converted to GGUF using:

llama.cpp

A tokenizer compatibility patch was applied during conversion to ensure successful GGUF export.

---

Credits

• Base Model: Mistral AI
• GGUF Conversion & Quantization: NANI-Nithin
• Tooling: llama.cpp