slots 5991, experts 48x512 = 24576 gen 0: 'Write a short story about a lighthouse keeper who finds a me' lookups 100171, distinct experts 10680 (1.78x cache) LRU cold 86.23% LFU cold 86.61% Belady (OPT) 89.34% pin 25% hot (other prompts) + LRU 83.83% pin 50% hot (other prompts) + LRU 80.42% pin 75% hot (other prompts) + LRU 76.07% LRU warm (same prompt, = bench run 2) 88.37% gen 1: 'Write a Python function that parses an ISO 8601 date string ' lookups 100517, distinct experts 11203 (1.87x cache) LRU cold 85.96% LFU cold 84.71% Belady (OPT) 88.85% LRU warm (prev prompt) 86.88% pin 25% hot (other prompts) + LRU 84.28% pin 50% hot (other prompts) + LRU 81.89% pin 75% hot (other prompts) + LRU 71.09% LRU warm (same prompt, = bench run 2) 87.41% gen 2: 'Explain the causes of the French Revolution to a high school' lookups 100109, distinct experts 12629 (2.11x cache) LRU cold 82.06% LFU cold 79.48% Belady (OPT) 87.38% LRU warm (prev prompt) 82.89% pin 25% hot (other prompts) + LRU 78.76% pin 50% hot (other prompts) + LRU 77.26% pin 75% hot (other prompts) + LRU 75.52% LRU warm (same prompt, = bench run 2) 83.40% gen 3: "Translate into French and explain the grammar: 'The children" lookups 100934, distinct experts 12377 (2.07x cache) LRU cold 81.30% LFU cold 80.13% Belady (OPT) 86.82% LRU warm (prev prompt) 82.48% pin 25% hot (other prompts) + LRU 79.42% pin 50% hot (other prompts) + LRU 76.25% pin 75% hot (other prompts) + LRU 72.00% LRU warm (same prompt, = bench run 2) 83.36% gen 4: 'Write a short story about a lighthouse keeper who finds a me' lookups 100171, distinct experts 10680 (1.78x cache) LRU cold 86.23% LFU cold 86.61% Belady (OPT) 89.34% LRU warm (prev prompt) 87.41% pin 25% hot (other prompts) + LRU 83.83% pin 50% hot (other prompts) + LRU 80.42% pin 75% hot (other prompts) + LRU 76.07% LRU warm (same prompt, = bench run 2) 88.37% story, cache-size sweep (slots: LRU / OPT) 2995 (0.5x): 72.81% / 84.68% 5991 (1.0x): 86.23% / 89.34% 8986 (1.5x): 89.21% / 89.34% 11982 (2.0x): 89.34% / 89.34%