Model Overview

Basic Information

• Model Name: Pythia-Chat
• Developer/Creator: Together AI
• Release Date: 2023
• Version: Base 7B v0.16
• Model Type: Transformer model

Description

Overview

The Pythia-Chat-Base-7B-v0.16 model is a 7 billion parameter language model developed by Together Computer. It is a fine-tuned version of EleutherAI's Pythia-7B model, with a focus on dialog-style interactions. The model is designed to assist developers in creating chatbots and conversational AI applications.

Key Features

• Fine-tuned on conversational data: The model is fine-tuned on the OIG dataset, which contains 43 million instructions, to improve its ability to engage in natural conversations.
• Supports multiple programming languages: Pythia-Chat-Base-7B-v0.16 can be used with a variety of programming languages, making it a versatile tool for developers.
• Quantized to 8-bit: The model is quantized to 8-bit, which reduces its memory footprint and allows it to run on GPUs with less memory, such as the Nvidia T4 with 16GB VRAM.

Intended Use

The Pythia-Chat-Base-7B-v0.16 model is intended for use in scenarios where developers need to create chatbots and conversational AI applications. It can be used for tasks such as:

• Customer support: The model can be used to create chatbots that provide customer support, answering frequently asked questions and guiding customers through various processes.
• Personal assistants: The model can be used to create personal assistants that can help users with tasks such as scheduling appointments, setting reminders, and providing information.
• Educational applications: The model can be used to create chatbots that can provide educational content and answer questions on various topics.

Language Support

The model supports multiple programming languages, including Python, Java, JavaScript, C++, and Go. It can be used with a variety of natural languages as well, although the primary language used in the training data is English.

Technical Details

Architecture

The Pythia-Chat-Base-7B-v0.16 model is based on the transformer architecture, with some modifications made by EleutherAI. It allows it to process and generate text efficiently.

Training Data

The model is fine-tuned on the OIG dataset, which contains 43 million instructions. The dataset was created by Together Computer in collaboration with LAION and Ontocord.ai. The model was further fine-tuned on user feedback submissions, which were released as the open-source together-user-feedback dataset.

Performance Metrics

The Pythia-Chat-Base-7B-v0.16 model has been evaluated on several benchmarks and has achieved strong results:

• Inference speed: The model can run on GPUs with as little as 16GB of memory, such as the Nvidia T4, making it suitable for real-time applications.
• Conversational ability: The model exhibits strong conversational ability, handling a variety of tasks and providing responses that are similar in quality to larger models such as GPT-NeoXT-Chat-Base-20B.

Usage

API Usage Example

Ethical Guidelines

Together Computer has focused on raising the bar for data governance and has been transparent about the data used to train the model. An opt-out process was provided for source code developers who did not want their code included in the dataset.

License Type

The Pythia-Chat-Base-7B-v0.16 model is licensed under the Apache 2.0 license, which allows for both commercial and non-commercial use of the model.