lunahr/Hermes-3-Llama-3.2-3B-abliterated

TEXT GENERATIONConcurrency Cost:1Model Size:3.2BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Dec 13, 2024License:llama3Architecture:Transformer0.0K Cold

Hermes 3 - Llama-3.2 3B (Abliterated) is a 3.2 billion parameter full parameter fine-tune of the Llama-3.2 foundation model by Nous Research, part of the Hermes series of LLMs. This model is optimized for advanced agentic capabilities, improved roleplaying, reasoning, multi-turn conversation, and long context coherence. It features powerful steering capabilities and control for the end user, excelling in generalist assistant tasks, function calling, structured output, and code generation.

Loading preview...

Hermes 3 - Llama-3.2 3B (Abliterated) Overview

Hermes 3 - Llama-3.2 3B is a 3.2 billion parameter language model developed by Nous Research, representing their first fine-tune in this parameter class within the Hermes series. It is a full parameter fine-tune of the Llama-3.2 foundation model, designed to provide powerful user steering and control.

Key Capabilities

  • Enhanced Generalist Performance: Offers significant improvements over Hermes 2 in agentic capabilities, roleplaying, reasoning, multi-turn conversation, and long context coherence.
  • Function Calling & Structured Output: Builds upon the Hermes 2 capabilities with more reliable function calling and structured output generation, including a specific prompt format for JSON mode.
  • Code Generation: Features improved skills in code generation.
  • ChatML Format: Utilizes the ChatML prompt format, enabling structured multi-turn dialogue and OpenAI API compatibility, with support for system prompts to guide model behavior.

Benchmarks

The model demonstrates competitive performance against Llama-3.1 Instruct models, with specific strengths and weaknesses. Benchmarks include:

  • GPT4All Average: 72.59
  • AGIEval Average: 44.05
  • BigBench Average: 44.13

Good For

  • Developers seeking a compact yet powerful LLM for agentic workflows.
  • Applications requiring robust function calling and structured JSON outputs.
  • Use cases benefiting from advanced roleplaying and multi-turn conversational abilities.
  • Scenarios where user control and steerability over the LLM's responses are crucial.

Finetunes
1 models