| title: D1337 CIPHER Training | |
| emoji: 🔥 | |
| colorFrom: red | |
| colorTo: purple | |
| sdk: docker | |
| pinned: true | |
| license: mit | |
| app_port: 7860 | |
| # D1337 CIPHER - Custom Training Environment | |
| **D1337 SOVEREIGN LABS** | |
| Custom QLoRA training environment for fine-tuning GLM-4.7-Flash-abliterated (31B) on cybersecurity datasets. | |
| ## Features | |
| - 🔥 **QLoRA Training** - Memory efficient training for 31B models | |
| - 🎯 **4x L40S Optimized** - Configured for 192GB VRAM | |
| - 📊 **Gradio UI** - Real-time monitoring and control | |
| - 🚀 **Auto Push to Hub** - Automatically saves to HuggingFace | |
| ## Configuration | |
| | Parameter | Value | | |
| |-----------|-------| | |
| | Base Model | `huihui-ai/Huihui-GLM-4.7-Flash-abliterated` | | |
| | Dataset | `Desorden1337/d1337-cipher-dataset` | | |
| | LoRA Rank | 64 | | |
| | LoRA Alpha | 128 | | |
| | Epochs | 5 | | |
| | Learning Rate | 2e-4 | | |
| | Max Seq Length | 4096 | | |
| ## Training Topics | |
| - SentinelOne EDR | |
| - CrowdStrike Falcon | |
| - Palo Alto Networks | |
| - Zero-day Research | |
| - AI/ML Security | |
| - Adversarial Attacks | |
| - Cloud Security | |
| ## Usage | |
| 1. Open this Space | |
| 2. Click "Start Training" | |
| 3. Monitor progress in logs | |
| 4. Model will be saved to `Desorden1337/d1337-cipher-v1` | |
| ## Hardware Requirements | |
| - **Minimum**: 4x L40S (192GB VRAM) | |
| - **Recommended**: 8x L40S or 4x A100 | |
| --- | |
| *D1337 SOVEREIGN LABS - Building the future of AI security* | |