InCoder-6B
| Property | Value |
|---|---|
| Model Size | 6 Billion parameters |
| License | CC-BY-NC 4.0 |
| Author | |
| Paper | Research Paper |
What is incoder-6B?
InCoder-6B is a sophisticated decoder-only Transformer model designed specifically for code generation and understanding. Developed by Facebook, this model represents a significant advancement in AI-powered coding assistance, trained on a vast dataset of permissively licensed open-source repositories from GitHub and GitLab, as well as StackOverflow content.
Implementation Details
The model is available in two versions: full-precision (float32) and half-precision (float16), making it adaptable to different computational requirements. It requires PyTorch, tokenizers (≥0.12.1), and transformers libraries for implementation. The model can run on a 16GB GPU for inference tasks when using the half-precision version.
- Supports 28+ programming languages with focus on Python and JavaScript
- Trained on Apache 2.0, MIT, BSD-2, and BSD-3 licensed codebases
- Implements causal-masked objective for code infilling
- Available in both full and half-precision versions
Core Capabilities
- Standard left-to-right code generation
- Code insertion and infilling
- Multi-language support
- Context-aware code completion
- Stack Overflow integration for practical coding solutions
Frequently Asked Questions
Q: What makes this model unique?
InCoder-6B stands out due to its ability to both generate code from left-to-right and perform code infilling, along with its extensive training on permissively licensed code repositories. The dual-precision availability makes it versatile for both training and inference tasks.
Q: What are the recommended use cases?
The model is ideal for code completion, generation, and understanding tasks across multiple programming languages. It's particularly well-suited for Python and JavaScript development, and can be used for both small-scale code suggestions and larger code generation tasks.
