InCoder-32B: Industrial Code Foundation Model
InCoder-32B (Industrial-Coder-32B) is the first 32B-parameter code foundation model purpose-built for industrial code intelligence. While general code LLMs excel at standard programming tasks, InCoder-32B is specifically designed to address challenges in hardware semantics, specialized language constructs, and strict resource constraints.
Model Description
InCoder-32B unifies code intelligence across several industrial engineering domains:
- Chip Design (Verilog / RTL)
- GPU Kernel Optimization (CUDA / Triton)
- Embedded Systems (ARM Cortex-M, STM32)
- Compiler Optimization (x86-64 assembly, LLVM)
- 3D Modeling (CAD/CAM via CadQuery / OpenCascade)
The model supports a native long-context window of up to 128K tokens.
Links
- Paper: InCoder-32B: Code Foundation Model for Industrial Scenarios
- GitHub: CSJianYang/Industrial-Coder
- Project Page: IndustrialCoder
Performance Highlights
InCoder-32B leads open-weight baselines across industrial domains and surpasses proprietary models like Claude-Sonnet-4.6 on specific benchmarks such as CAD-Coder IoU and KernelBench.
| Domain | Benchmark | InCoder-32B | Claude-Sonnet-4.6 |
|---|---|---|---|
| Chip Design | RealBench Func@1 (Mod) | 62.7 | 37.2 |
| GPU Optim. | KernelBench L1/L2/L3 | 22.2/36.0/14.0 | 11.1/28.0/2.0 |
| 3D Modeling | CAD-Coder Compile (%) | 82.0 | 77.0 |
| Code Optim. | SuperCoder Acc | 91.0 | 88.0 |
Quickstart
Installation
pip install -U "transformers>=4.57.1" accelerate safetensors
Usage with Transformers
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "Multilingual-Multimodal-NLP/IndustrialCoder"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto",
trust_remote_code=True,
)
messages = [{"role": "user", "content": "Optimize this CUDA kernel for better memory coalescing."}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)
with torch.no_grad():
out = model.generate(**inputs, max_new_tokens=2048, temperature=0.6, top_p=0.85, top_k=20)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
Training Pipeline
The model is trained via a three-stage Code-Flow pipeline:
- Pre-training & Annealing: General pre-training followed by curated industrial code annealing.
- Mid-training: Progressive context extension from 8K to 128K tokens using synthetic industrial reasoning data.
- Post-training: Execution-grounded SFT with 2.5M samples and feedback-driven repair trajectories.
Citation
@article{yang2025incoder32b,
title={InCoder-32B: Code Foundation Model for Industrial Scenarios},
author={Yang, Jian and Zhang, Wei and Wu, Jiajun and Cheng, Junhang and Guo, Shawn and Wang, Haowen and Gu, Weicheng and Du, Yaxin and Li, Joseph and Xu, Fanglin and others},
journal={arXiv preprint arXiv:2603.16790},
year={2025}
}
Disclaimer
The model may generate incorrect or unsafe code. Always review and test outputs in a sandboxed environment before production use. Industrial code (RTL, embedded firmware, GPU kernels) requires expert human review before deployment.