metadata

license: apache-2.0

Model Card for Model ID

bling-answer-tool is a quantized version of BLING Tiny-Llama 1B, with 4_K_M GGUF quantization, providing a very fast, very small inference implementation for use on CPUs.

bling-tiny-llama is a fact-based question-answering model, optimized for complex business documents.

To pull the model via API:

from huggingface_hub import snapshot_download           
snapshot_download("llmware/bling-answer-tool", local_dir="/path/on/your/machine/", local_dir_use_symlinks=False)

Load in your favorite GGUF inference engine, or try with llmware as follows:

from llmware.models import ModelCatalog  
model = ModelCatalog().load_model("bling-answer-tool")            
response = model.inference(query, add_context=text_sample)

Note: please review config.json in the repository for prompt wrapping information, details on the model, and full test set.

Model Description

Developed by: llmware
Model type: GGUF
Language(s) (NLP): English
License: Apache 2.0
Quantized from model: llmware/bling-tiny-llama

Model Card Contact

Darren Oberst & llmware team