nicoboss/dolphin-2.9.3-mistral-nemo-12b-llamacppfixed

Dolphin 2.9.3 Mistral Nemo 12b Overview

Dolphin 2.9.3 Mistral Nemo 12b is a 12 billion parameter language model developed by Eric Hartford and Cognitive Computations, built upon the mistralai/Mistral-Nemo-Base-2407 architecture. It supports a substantial context length of 32768 tokens, though its finetuning utilized an 8192 sequence length. The model employs the ChatML prompt template format.

Key Capabilities

• Instruction Following & Conversation: Excels in general instruction adherence and engaging in conversational exchanges.
• Coding Skills: Demonstrates proficiency in coding tasks, including code generation and translation, as evidenced by its training datasets.
• Agentic Abilities: Possesses initial capabilities for agentic workflows and supports function calling, enabling more complex interactions.
• Uncensored Nature: The model's training data was filtered to remove alignment and bias, resulting in an uncensored model that is highly compliant with user requests, including potentially unethical ones. Users are advised to implement their own alignment layers.

Training Details

The model was trained using Axolotl on a diverse set of ShareGPT-formatted datasets, including those focused on coding (e.g., dolphin-coder-translate, dolphin-coder-codegen), mathematical reasoning (Orca-Math), and agentic instruction following (agent_instruct_react, toolbench_instruct). Training involved 3 epochs with a learning rate of 5e-6 and a total batch size of 128 across 8 GPUs. The base model has a 128K context, while finetuning used 8192 sequence length.

Good For

• Developers requiring a highly compliant and uncensored model for various tasks.
• Applications needing strong instruction following, conversational, and coding capabilities.
• Experimentation with agentic AI and function calling in a 12B parameter model.