Open LLM List
Tiny Chat 1776 K
raincandy-uIntroduction
TINYCHAT-1776K is a lightweight language model developed by raincandy-u, specifically trained on the TinyChat dataset. The model aims to generate natural responses using a minimal architecture suitable for basic conversational scenarios. It is tailored to handle English at a level appropriate for three-year-old children and lacks broader world knowledge.
Architecture
The model is based on the llama architecture and is configured with the following specifications:
- Hidden Size: 192
- Intermediate Size: 640
- Number of Attention Heads: 16
- Number of Hidden Layers: 3
- Number of Key Value Heads: 4
- Vocabulary Size: 2048
- Max Position Embeddings: 256
- Tie Word Embeddings: True
Training
The TINYCHAT-1776K model is trained from scratch using conversations typical of young children, focusing on generating simple and coherent responses. Due to its limited scope, the model is not suitable for tasks requiring extensive knowledge or complex reasoning.
Guide: Running Locally
To run the TINYCHAT-1776K model locally, follow these steps:
- Clone the Repository: Download the model files from Hugging Face.
- Install Dependencies: Ensure you have the required libraries, such as
transformers, installed in your environment. - Load the Model: Use the
AutoConfigandAutoModelclasses to load and configure the model. - Inference: Set generation parameters (e.g.,
top_k=40,top_p=0.8,temperature=1) and run the model to generate responses.
For better performance, consider using cloud GPUs from providers such as AWS, Google Cloud, or Azure.
License
The TINYCHAT-1776K model is licensed under the Apache-2.0 license, which allows for both personal and commercial use, modifications, and distribution.
