October 2025 monthly summary focusing on key accomplishments, business value, and technical achievements across basetenlabs/truss-examples and basetenlabs/truss. Highlights include deployment examples for Briton Inference Stack v2 with FP8 configurations, a rollback to stabilize the release, FP4_MLP_ONLY quantization support and a Truss rc4 version bump for improved release visibility. These efforts delivered tangible business value by enabling optimized deployment options, maintaining stability, expanding hardware-accelerator compatibility, and improving release traceability.

During Sep 2025, delivered FP4-quantized model deployment capabilities and related documentation across two repositories to broaden deployment options and reduce compute needs. Specifically, FP4 deployment examples and docs for embeddings, reranking, and Llama/Qwen models were added to basetenlabs/truss-examples, with README and YAML updates to guide users through FP4 deployments. In basetenlabs/truss, FP4_KV quantization support was integrated into the configuration and validation logic (trt_llm_config.py), enabling FP4_KV usage alongside FP8 context FMHA, with a package version bump to reflect changes. A packaging/versioning fix aligned pyproject.toml and uv.lock to the correct 0.11.8rc4 revision to ensure accurate version tracking.