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 506: 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 293: 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 729: 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 937: 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 91: 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 747: 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 592: 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 934: 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 192: 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 156: 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 446: 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 596: 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 369: 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 659: 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 956: 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 834: 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 194: 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 886: 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 836: 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 366: 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 438: 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 703: 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 64: 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 142: 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 994: 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 777: 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 350: 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 997: 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 166: 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 135: 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 491: 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 860: 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 231: 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 30: 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 669: 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 488: 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 697: 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 693: 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 12: 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 723: 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 732: 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 178: 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 763: 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 795: 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 248: 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 529: 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 345: 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 495: 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 174: 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 334: 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 263: 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 887: 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 295: 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 721: 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 214: 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 1: 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 358: 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 560: 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 127: 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 530: 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 56: 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 786: 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 764: 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 359: 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 620: 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 705: 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 462: 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 39: 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 762: 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 468: 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 557: 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 95: 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 610: 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 221: 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 110: 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 105: 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 66: 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 490: 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 604: 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 674: 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 329: 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 306: 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 457: 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 423: 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 797: 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 992: 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 206: 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 177: 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 74: 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 26: 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 809: 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 492: 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 930: 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 185: 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 554: 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 559: 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 161: 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 657: 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 346: 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 424: 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.