Models

Notice: After Quick Start, all these models below are all installed. You should NOT download them again.

Text Generation Model

Overview

Text generation is the main model used for communicating between user and LocalAssistant. Got labeled as 'Text_Generation' or '1'.

Downloading

I recommend to try Qwen, no need for hf_token and is light-weight.

locas.cmd download -n qwen Qwen/Qwen2.5-1.5B-Instruct 1

(Choose other models from here.)

If you're too familiar with HuggingFace and confident enough about your device, try 'meta-llama/Meta-Llama-3-8B'.

Sentence Transformer Model

Overview

Sentence transformer is a model used to semantic search data from provided documents. Required for locas docs .... Got labeled as 'Sentence_Transformer' or '2'.

Downloading

What this model does is encoding docs and retrieving information. You will NOT need any fancy and advance models like MsMarco.

locas.cmd download -n minilm sentence-transformers/all-MiniLM-L6-v2 2

(Choose other models from here.)

Cross Encoder Model

Overview

Cross encoder is a model used to rerank data after retrieving information process. Required for locas docs .... Got labeled as 'Cross_Encoder' or '3'.

Download

locas.cmd download -n msmarco cross-encoder/ms-marco-MiniLM-L-6-v2 3

(Choose other models from here.)

Built-in Model

Overview

Models here is NOT trained by me, but rather being used for LocalAssistant's function. If one is broken by any chances, try remove its folder in models/built-in and run installation again.

Rebel - Relation Extraction

Source: Babelscape/rebel-large.

This model is used for relation extraction and creating knowledge graph. Required for memory function in locas start and relation extraction in locas docs extract.