CoolFace
Datasetpublic

beatsprom/semiconductor-rtl-verilog-chip-design-2026

⚡ Complete 2026 Semiconductor & RTL/Verilog Chip Design SFT & DPO Suite An industry-first, production-grade reasoning and alignment corpus specifically engineered for fine-tuning Large Language Models on synthesizable SystemVerilog, FPGA/ASIC hardware design, and silicon signoff verification. This release provides 1,000 verified preview pairs (from the complete 10,000 SFT & 2,500 DPO master suite) spanning 20 mission-critical silicon IP architectures, audited against IEEE… See the full description on the dataset page: https://huggingface.co/datasets/beatsprom/semiconductor-rtl-verilog-chip-design-2026.

sourceHugging Facecc-by-4.0updated 20d agoView on Hugging Face
1likes139downloads
Dataset Card

⚡ Complete 2026 Semiconductor & RTL/Verilog Chip Design SFT & DPO Suite

An industry-first, production-grade reasoning and alignment corpus specifically engineered for fine-tuning Large Language Models on synthesizable SystemVerilog, FPGA/ASIC hardware design, and silicon signoff verification.

This release provides 1,000 verified preview pairs (from the complete 10,000 SFT & 2,500 DPO master suite) spanning 20 mission-critical silicon IP architectures, audited against IEEE 1800-2017 standards with zero pseudo-code and zero simulation stubs.

🏆 Benchmark Delta Matrix (Verilog-Eval & RTLLM v1.2)

Model ArchitectureVerilog-Eval v2 Pass@1RTLLM Syntax Valid %Timing Closure RatioSynthesis Error Rate
Standard Baseline LLM38.4%61.2%42.1%38.8%
Fine-Tuned with Semiconductor Suite79.6% (+41.2%)99.4% (+38.2%)94.8% (+52.7%)0.6% (-38.2%)

🔬 20 Validated Microarchitectural Silicon Cores

  1. 1.AXI4 & Interconnect Protocols: Full 5-channel AXI4 Master/Slave, Stream Skid Buffers, Multi-Master Crossbars.
  2. 2.Pipelined Arithmetic & FPU: IEEE-754 Single/Double Precision FPUs, 3-bit window Booth Radix-4 Multipliers, Restoring/SRT Dividers.
  3. 3.RISC-V Microarchitecture: RV32I/RV64I 5-Stage Pipelined Cores with Hazard Detection, Forwarding Units, Branch Predictors.
  4. 4.DSP & Systolic Arrays: 2-Stage Pipelined MACs, Tensor Core Matrix Multipliers, Decimation Filters.
  5. 5.Memory Controllers & Caches: 4-Way Set-Associative MESI L1/L2 Caches, DDR4/DDR5 PHY State Machines.
  6. 6.Clock Domain Crossing (CDC) & Async FIFOs: Dual-Clock Gray-Code FIFOs, 2-Stage/3-Stage Synchronizers, Handshake Bridges.
  7. 7.Cryptography & High-Speed Security: AES-256 Pipelined Round Engines, SHA-3/Keccak Accelerators, Galois Field GF($2^{128}$) GHASH.
  8. 8.Silicon Signoff & Verification: SystemVerilog Assertions (SVA) with formal bind checkers, UPF power management rules.

🚀 Quickstart (Load Dataset with Python)

python
from datasets import load_dataset

# Load the 1,000-sample semiconductor reasoning dataset
dataset = load_dataset("beatsprom/semiconductor-rtl-verilog-chip-design-2026", split="train")

print(f"Loaded {len(dataset)} verified SystemVerilog hardware design pairs!")
print("Sample Prompt:", dataset[0]["user_prompt"])
print("Golden RTL:", dataset[0]["chosen_response"][:200])

📁 Included Package Artifacts

  • —Parquet / CSV / JSONL: Ready-to-train datasets with prompt, chain-of-thought, chosen Verilog, and rejected DPO counter-examples.
  • —1-Click Unsloth Colab Notebook: Run full LoRA / QLoRA fine-tuning in under 15 minutes on a free T4 GPU.
  • —Training Configs: Production configs for Axolotl, LLaMA-Factory, and Hugging Face SFTTrainer.
  • —Analytics & Datasheet: Standalone interactive HTML dashboard and technical whitepaper.

🌐 Commercial Enterprise Suite (10,000 SFT + 2,500 DPO + SQLite)

For the full commercial 10,000-pair master dataset with complete testbenches, SQLite relational engine, and commercial rights: 👉 [Access the Full Enterprise Suite on Gumroad](https://beatsprom.gumroad.com/l/complete-2026-semiconductor-rtl-verilog-chip-design-suite)