April 2026 monthly summary for huggingface/transformers: Key feature delivered was the Flexible data type configuration for the Four Over Six quantization method, enabling customizable dtype settings to improve training flexibility with quantized tensors. No major bug fixes were documented for this repo in April. The change lays groundwork for more robust quantization workflows and can improve model training efficiency in quantized environments.

Month: 2025-05 | Repository: neuralmagic/guidellm. Focused on enhancing the OpenAIHTTPBackend to improve reliability and flexibility by adding redirect handling and support for extra query parameters. These changes enable follow_redirects and allow passing custom parameters to OpenAI endpoints (chat completions, models, text completions). No high-severity bugs reported this month; main achievements centered on feature delivery with clear business value.