Silicon-DPO Platinum: The Reasoning-Code Dataset

INDUSTRIAL-GRADE REASONING DATA.

Most datasets are static. Ours is compiled.

Silicon-DPO represents the shift from "LLM-as-a-judge" to "Runtime-Verification."

### 🔬 The Spec

* Volume: 5,001 High-Density Pairs

* Verification: TDD Sandbox (Python Unit Tests)

* Format: .jsonl (Compatible with Axolotl, Llama-Factory)

* Structure: <think> Trace + Self-Corrected Code

* Domain: Coding (35%), Finance (20%), STEM (15%)

### 🛡️ Why Silicon-DPO?

We use an agentic loop (The Titan Architecture) to generate code, execute it, debug failures, and record the repair process.

You get the "Self-Healing" traces that define modern reasoning models (like o1).

---

### 💎 Licensing Tiers

Option A: Personal / Research (€49)

* For academic research and individual hobbyists.

* Strictly Non-Commercial.

Option B: Commercial / Startup (€499)

* Full rights to train and deploy proprietary models.

* Royalty-free usage for your end products.