Pinned Loading

1. llm-inference-performance llm-inference-performance Public

   Serving LLMs on H100s with NVIDIA Dynamo, vLLM and Kubernetes: first-principles performance analysis in seven hands-on exercises

   Python 1

2. vllm-internals-viz vllm-internals-viz Public

   Interactive replay of vLLM's V1 engine loop: TypeScript port of the scheduler, KV cache manager and block pool, verified by differential replay against real vLLM traces

   TypeScript 1

3. llm_eval llm_eval Public

   Evaluation of LLM powered systems.

   Python 1

4. llm-d-lambda-deployment llm-d-lambda-deployment Public

   llm-d feature testing — aggregated inference (prefix-cache routing, queue-depth balancing, HPA autoscaling) and P/D disaggregated inference (NIXL KV transfer on time-sliced GPU) on NVIDIA GH200

   Shell 1

5. moe-disaggregation-benchmark moe-disaggregation-benchmark Public

   Does prefill/decode disaggregation pay for an MoE model? SLO-first goodput benchmark of Qwen3-30B-A3B on 4x H100 and 8x A100 with vLLM, llm-d and NVIDIA Dynamo, with per-request data.

   Python

6. moe-agentic-serving-benchmark moe-agentic-serving-benchmark Public

   Benchmark: how many agentic sessions one 8x H100 node serves on Qwen3-235B-A22B FP8 with NVIDIA Dynamo, vLLM and TensorRT-LLM, and what moves the limit.

   Python