AI BriefingReplit's New AI Model Now Available on Hugging Face
Key point
Replit has released its new code generation model, Replit Code V1.5 3B, on Hugging Face.
Details
Replit has released a new code generation language model, Replit Code V1.5 3B, on Hugging Face. This model release came alongside the announcement that Replit AI would be made available for free to all users.
The model is designed as a foundation model for fine-tuning toward specific applications, with fewer restrictions on commercial use. It was trained on 1 trillion tokens, including data from the Stack dataset and public data from StackExchange.
Key features include:
- Best-in-class HumanEval and Multi-PLe performance among 3B-scale models
- Support for 30+ programming languages via a 32K vocabulary
- Application of cutting-edge techniques such as Grouped Query Attention, Flash Attention Triton Kernels, and ALiBi
- High-quality data secured through code quality filtering and removal of harmful content
Notably, when fine-tuned on Replit user code, it outperforms larger models such as CodeLlama 7B. Users can immediately leverage the model for code completion tasks via the transformers library.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Qwen3.5-based Small Decision Model 'Kev' Released with TypeSafe API Compatibility and Performance Metrics