LLM Reference

WizardCoder 3B

About

WizardCoder 3B is an advanced large language model specializing in code-related tasks, based on the StarCoder architecture, a decoder-only transformer akin to LLaMA. It comprises 3 billion parameters and was trained on an extensive dataset from open-source code repositories and mathematical domains. This model excels in code generation, task completion, and offers a "Fill in the Middle" (FIM) feature. Despite its strong performance on benchmarks like HumanEval and MBPP, there are potential limitations such as biases from its training data and task-specific performance variations. Additionally, a more powerful variant, WizardCoder-33B-V1.1, is available, enhancing its capabilities further 236.

Capabilities

MultimodalFunction CallingTool UseJSON Mode

Specifications

Parameters3B
ArchitectureDecoder Only
Specializationgeneral