DeciCoder 1B

About

DeciCoder 1B is a 1 billion-parameter, open-source large language model focused on code generation 12. It excels at efficient and accurate code completion for Python, Java, and JavaScript using a unique Grouped Query Attention architecture with a 2048-token context window 13. Trained on a substantial dataset with a Fill-in-the-Middle objective, it offers impressive throughput, especially when used with Deci's Infery LLM inference engine 79. Although capable of single or multi-line code completion, it may produce suboptimal results as it's not an instruction-following model. Performance is benchmarked on HumanEval, showing variable accuracy across languages, and it is available under the Apache 2.0 license 25.

Capabilities

Providers(1)