bling-tiny-llama-onnx

bling-tiny-llama-onnx is a very small, very fast fact-based question-answering model, designed for retrieval augmented generation (RAG) with complex business documents, quantized and packaged in ONNX int4 for AI PCs using Intel GPU, CPU and NPU.

This model is one of the smallest and fastest in the series. For higher accuracy, look at larger models in the BLING/DRAGON series.

Model Description

• Developed by: llmware
• Model type: tinyllama
• Parameters: 1.1 billion
• Quantization: int4
• Model Parent: llmware/bling-tiny-llama-v0
• Language(s) (NLP): English
• License: Apache 2.0
• Uses: Fact-based question-answering, RAG
• RAG Benchmark Accuracy Score: 86.5

Model Card Contact

llmware on github
llmware on hf
llmware website