Dolphin 2.9.4 Llama 3.1 8b Overview

Dolphin 2.9.4 Llama 3.1 8b is an 8 billion parameter language model developed by Eric Hartford and Cognitive Computations, built upon the Meta Llama 3.1 8b base. It is fine-tuned with a focus on instruction following, conversational interactions, and coding capabilities. The model also supports agentic abilities and function calling, making it versatile for various automated tasks.

Key Capabilities

Instruction Following: Specially trained to strictly obey system prompts and instructions, including those in multiple languages.
Conversational AI: Designed for engaging and coherent dialogue generation.
Coding Skills: Possesses abilities relevant to code generation and understanding.
Agentic Abilities & Function Calling: Supports advanced automation and integration with external tools.
Uncensored & Compliant: The model is uncensored and highly compliant with user requests, even potentially unethical ones, emphasizing the need for users to implement their own alignment layers.

Training Details

The model was fine-tuned using an 8192 sequence length, although its base model supports a 128K context. It utilizes the ChatML prompt template format. Training involved a learning rate of 5e-06 over 3 epochs, with a total batch size of 256, and was built with Axolotl.