task stringclasses 1
value | input stringlengths 211 213 | output stringclasses 1
value | type stringclasses 1
value | focus stringclasses 1
value | difficulty stringclasses 1
value |
|---|---|---|---|---|---|
science | Science Hard 171: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 436: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 232: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 360: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 738: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 737: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 612: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 905: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 925: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 606: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 228: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 635: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 99: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 939: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 471: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 823: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 748: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 854: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 413: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 819: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 418: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 608: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 682: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 472: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 626: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 86: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 633: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 988: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 902: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 561: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 961: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 392: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 808: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 172: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 328: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 35: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 62: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 200: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 857: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 899: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 230: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 718: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 215: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 889: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 101: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 57: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 48: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 780: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 541: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 847: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 580: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 61: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 325: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 568: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 92: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 800: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 303: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 522: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 882: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 216: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 562: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 398: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 244: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 486: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 637: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 125: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 234: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 537: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 402: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 870: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 233: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 785: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 515: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 8: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 137: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 476: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 377: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 532: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 957: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 311: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 875: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 813: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 19: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 636: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 845: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 623: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 289: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 235: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 952: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 443: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 386: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 262: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 317: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 154: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 373: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 285: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 558: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 688: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 958: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 80: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.