llama-3.2-3b-onnx-qnn

llama-3.2-3b-onnx-qnn is an ONNX QNN int4 quantized version of Llama 3.2 3B Instruct, providing a very small, very fast inference implementation, optimized for NPU deployment on Windows ARM64 AI PCs with Snapdragon Elite X NPU processors.

llama-3.2-3b-instruct is a new 3B chat foundation model from Meta.

Model Description

• Developed by: meta-llama
• Quantized by: llmware
• Model type: llama-3.2
• Parameters: 3 billion
• Model Parent: meta-llama/Meta-Llama-3.2-3B-Instruct
• Language(s) (NLP): English
• License: Llama 3.2 Community License
• Uses: General chat use cases
• RAG Benchmark Accuracy Score: NA
• Quantization: int4

Model Card Contact

llmware on github

llmware on hf

llmware website