Optuna

tuner.py is DigitalOcean's optimization agent for finding vLLM serving configurations for a given model. It uses Bayesian optimization through Optuna to propose vLLM and Kubernetes settings, deploy trials on inference hardware, measure throughput under latency targets, and retain the best configuration or restore the previous setup. It produces an HTML results report and an optimized YAML serving recipe; the videos describe experiments involving GPT-OSS and DeepSeek R1 Distill, as well as planned extensions for disaggregated prefill and decode serving.

OPAL (OPtimizer with an Agent in the Loop) is a PyTorch optimizer wrapper developed by Xingjian Bai and contributors. It turns a fixed optimizer into two editable programs: an updater that converts gradients into parameter updates, and an analysis program that reports training statistics. A coding agent, Claude Code by default, periodically reads the goal, losses, gradients, and analysis reports, then can rewrite learning rates and schedules, momentum, preconditioning, clipping, parameter-group update rules, and related optimizer logic while training continues. New programs are trialed for 20 steps and can be reverted on errors when a parameter-and-state snapshot is available; until a change is made, the wrapped run is bit-for-bit equivalent to the original optimizer run. OPAL supports CPU, single-GPU, and single-node DDP PyTorch training, with integrations for selected Hugging Face Trainer and Accelerate workflows, but does not support multi-node training, FSDP, or DeepSpeed. Agent runs require Claude Code and an authenticated account, while the keep backend provides an agent-free dry run. The project is distributed under the MIT license.

Optima is an Artificial Analysis product for creating custom benchmarks from user-supplied context. An Optima agent uses that context to evaluate models and produce tradeoff curves comparing task-specific intelligence with cost.

NVIDIA Model Optimizer (ModelOpt) is an open-source library for optimizing deep-learning and generative-AI models for inference deployment. It accepts Hugging Face, PyTorch, or ONNX models and provides Python APIs for composing post-training quantization, quantization-aware training and distillation, pruning, neural architecture search, speculative decoding, and sparsity techniques, producing optimized or quantized checkpoints. The resulting checkpoints can be exported for inference frameworks including TensorRT-LLM, TensorRT, vLLM, and SGLang. The library also integrates with NVIDIA Megatron-Bridge, Megatron-LM, Hugging Face Accelerate, Transformers, and Diffusers for supported training and export workflows. It is distributed as the nvidia-modelopt package on PyPI and can also be installed from the NVIDIA GitHub repository or used through NVIDIA container images.

Optiver is a global, technology- and research-driven market-making and trading firm. Its teams of engineers, scientists, mathematicians, quantitative researchers and traders develop trading systems, models and strategies across markets, products and time zones. The company processes petabytes of market data daily, executes more than 10 million trades per day, connects to more than 100 exchanges and prices more than a million financial instruments, according to its website. Optiver's internal AI platform is designed for rapid, measurable deployment in trading workflows. It includes an inference gateway for routing and budget control, internal golden sets for benchmarking new frontier models, and a model- and harness-agnostic meta harness; the platform runs on premises so changes can reach traders and their P&L impact can be evaluated within hours.