Sean Li

Qwen3.5-35B-A3B-FP8 on NVIDIA GB10 Blackwell - 262K context, 53 tok/s, MTP

Functionary v3.2 tool router for CCA — ARM64+CUDA (DGX Spark GB10)

vLLM for NVIDIA GB10 DGX Spark - SM121 Blackwell, FlashInfer + Triton MoE

Qwen3.5-9B-FP8 with thinking + tool calling — CCA note-taker and tool orchestrator.

Qwen3-Embedding-8B via vLLM pooling — 4096-dim embeddings for CCA code intelligence.

LLM inference benchmark tool with web dashboard for OpenAI-compatible APIs