LLM Reference

Replit Code

About

Replit Code is a series of large language models (LLMs) focused on code completion and generation, such as Replit Code v1.3B and v1.5 3B. These models are trained on extensive datasets featuring various programming languages, including Python, JavaScript, and Java 56. They utilize advancements like Flash Attention and AliBi positional embeddings to enhance training and inference efficiency 56. Capabilities encompass code completion, generation from natural language prompts, and support for multiple languages 5. These models are integrated into Replit's platform for code suggestions and explanation tools, initially offered through paid services but now with basic AI features available for free 12. They are also accessible on Hugging Face, inviting community participation and development 10. Performance varies by model version and language used 78.

Capabilities

VisionMultimodalReasoningFunction CallingTool UseJSON ModeCode Execution

Providers(1)

ProviderInput (per 1M)Output (per 1M)Type
Replicate APIServerless

Rankings

Specifications

Released2023-06-15
Parameters3B
ArchitectureDecoder Only
Specializationgeneral
Trainingfinetuning

Providers(1)