RAG recall vs speed tradeoff

[localpatch]

Appreciate the detail. I ended up in about the same place using vLLM. I got better results once I stopped chasing one metric and just looked at whethe

worth checking the driver first.

[fastbench]

What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Mistral and compare from there. Once you know the plain path works, then add the faster path back in piece by piece.

[vectorpoint]

Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal.

[plainworks]

Not sure I agree there. I get why that seems right, but on GB200 the bottleneck felt elsewhere.