LLM Reference

OpenChat 3.2 Super

About

OpenChat 3.2 Super is a robust open-source large language model with 13 billion parameters, crafted on the Llama 2 framework for superior conversational AI and instruction-following capabilities. It stands out by securing top positions on benchmarks like AgentBench and AlpacaEval, while excelling in tests like MT-bench with a score of 7.19. Trained on a dataset of approximately 80,000 ShareGPT conversations, the model impresses with an efficient fine-tuning process that utilizes a conditioning strategy and weighted loss, taking just 15 hours on 8 A100 80G GPUs. Although it shows strong performance, challenges remain in areas such as complex reasoning and programming, occasionally producing inaccurate information. OpenChat 3.2 Super is commercially available under the Llama 2 Community License.

Capabilities

MultimodalFunction CallingTool UseJSON Mode

Specifications

Parameters13B
Context4K
ArchitectureDecoder Only
Specializationgeneral