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 421: 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 909: 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 225: 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 696: 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 924: 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 811: 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 397: 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 832: 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 464: 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 261: 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 160: 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 63: 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 3: 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 740: 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 430: 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 385: 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 855: 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 798: 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 197: 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 579: 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 527: 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 825: 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 965: 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 278: 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 280: 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 432: 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 545: 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 352: 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 601: 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 461: 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 29: 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 217: 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 131: 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 796: 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 361: 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 910: 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 969: 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 81: 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 372: 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 888: 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 175: 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 403: 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 695: 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 209: 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 267: 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 5: 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 945: 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 90: 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 993: 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 14: 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 829: 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 959: 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 652: 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 663: 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 784: 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 183: 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 188: 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 750: 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 41: 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 130: 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 982: 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 765: 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 265: 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 727: 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 439: 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 321: 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 331: 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 356: 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 27: 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 104: 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 301: 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 313: 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 286: 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 670: 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 665: 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 660: 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 713: 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 650: 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 406: 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 211: 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 806: 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 378: 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 712: 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 818: 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 190: 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 614: 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 776: 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 28: 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 314: 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 837: 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 884: 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 456: 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 941: 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 728: 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 770: 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 15: 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 544: 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 494: 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 919: 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 33: 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.