nanogpt-training

• Analyzed: Apr 14, 2026, 1:01 PM
• Findings: 0
• Version: 0.1.0
• Hash: 1f1a024b3c5af46cafecc582e6e456a0e79538c221eb6deea4e714f1ec364c61
• Source: palm
• External report: View on VirusTotal

Overview

Type: OpenClaw Skill Name: mhc-layer-impl-nanogpt-training Version: 0.1.0 The skill bundle provides a comprehensive and legitimate implementation for training GPT-2 scale models using PyTorch and Modal. The code follows standard practices from well-known community projects (e.g., Karpathy's nanoGPT), including implementations for the Muon optimizer, Rotary Positional Embeddings, and memory-mapped data loading from HuggingFace Hub. No evidence of data exfiltration, malicious execution, or prompt injection was found; all network activity is limited to downloading public datasets from HuggingFace.