waqasm86/llcuda-models

llcuda Models

Optimized GGUF models for llcuda - Zero-config CUDA-accelerated LLM inference.

Models

google_gemma-3-1b-it-Q4_K_M.gguf

• Model: Google Gemma 3 1B Instruct
• Quantization: Q4_K_M (4-bit)
• Size: 769 MB
• Use case: General-purpose chat, Q&A, code assistance
• Recommended for: 1GB+ VRAM GPUs

Performance:

• Tesla T4 (Colab/Kaggle): ~15 tok/s
• Tesla P100 (Colab): ~18 tok/s
• GeForce 940M (1GB): ~15 tok/s
• RTX 30xx/40xx: ~25+ tok/s

Usage

With llcuda (Recommended)

pip install llcuda

import llcuda
engine = llcuda.InferenceEngine()
engine.load_model("gemma-3-1b-Q4_K_M")
result = engine.infer("What is AI?")
print(result.text)

With llama.cpp

# Download model
huggingface-cli download waqasm86/llcuda-models google_gemma-3-1b-it-Q4_K_M.gguf --local-dir ./models

# Run with llama.cpp
./llama-server -m ./models/google_gemma-3-1b-it-Q4_K_M.gguf -ngl 26

Supported Platforms

• ✅ Google Colab (T4, P100, V100, A100)
• ✅ Kaggle (Tesla T4)
• ✅ Local GPUs (GeForce, RTX, Tesla)
• ✅ All NVIDIA GPUs with compute capability 5.0+

Links

• PyPI: pypi.org/project/llcuda
• GitHub: github.com/waqasm86/llcuda
• Documentation: waqasm86.github.io

License

Apache 2.0 - Models are provided as-is for educational and research purposes.

Credits

• Model: Google Gemma 3 1B
• Quantization: llama.cpp GGUF format
• Package: llcuda by Waqas Muhammad