CYBERBRAIN - Phase 2 (Systems & Memory Safety)

Precision: Pure FP32 training (zero NaN/overflow drift).
Convergence: Training loss smoothly reduced from 1.05 to 0.34.
Full Backup: Download CYBERBRAIN_PHASE2_25M_FULL_PACKAGE.zip from the Files tab for the raw corpus, tokenized blocks, and model weights.

Continual pre-trained version of Qwen2.5-Coder-0.5B on system-level C and Rust codebases (concurrency, Arc/Mutex patterns, and pointer safety).