i-Coder/iCoder-27B
i-Coder/iCoder-27B is a 27 billion parameter model developed by i-Coder for industrial coding tasks, specifically RTL design and GPU kernel optimization. This model was developed through an AI-led process, where an agent coordinated a multi-stage pipeline including supervised fine-tuning and reinforcement learning. Despite its compact size, iCoder-27B demonstrates strong performance, leading on RTLLM with 68.0 and achieving the highest KernelBench L1 correctness at 61%, often surpassing much larger models in specialized coding benchmarks.
Loading preview...
iCoder-27B: AI-Led Industrial Coding Model
iCoder-27B is a 27-billion parameter language model developed by i-Coder, specifically engineered for industrial coding applications such as RTL (Register-Transfer Level) design and GPU kernel optimization. This model is notable for its unique development process, which involved an AI agent autonomously coordinating a multi-stage training pipeline. This pipeline included supervised fine-tuning, on-policy self-distillation, and reinforcement learning, with rewards derived directly from compiling and running the model's generated code.
Key Capabilities & Performance
- Specialized Industrial Coding: Excels in niche but critical areas of hardware description and GPU programming.
- Compact yet Powerful: Despite its 27B parameters, it outperforms models up to 59 times larger on specific benchmarks, including DeepSeek-V4-Pro, GLM-5.2, and Kimi-K2.6.
- Leading Benchmarks: Achieves a leading score of 68.0 on RTLLM (Functional avg@4), ties Claude Opus 4.8 for the best TritonBench-G pass@1 (20.1%), and boasts the highest KernelBench L1 correctness at 61% among evaluated models.
- AI-Driven Development: Showcases an innovative approach to model creation, where AI agents manage and optimize the training strategy.
When to Use This Model
iCoder-27B is ideal for developers and researchers focused on:
- Hardware Description Languages (HDL): Generating and optimizing Verilog or similar RTL code.
- GPU Kernel Development: Creating and optimizing high-performance GPU kernels.
- Exploring AI-Led Model Development: As a case study or foundation for further research into autonomous model training.
This model is particularly suited for tasks requiring high accuracy and functional correctness in specialized industrial coding domains, offering a powerful solution in a relatively efficient parameter count.