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 70: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 673: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 47: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 590: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 531: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 298: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 264: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 454: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 291: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 867: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 756: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 801: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 153: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 676: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 752: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 184: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 978: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 410: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 877: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 566: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 429: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 316: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 400: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 766: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 903: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 237: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 399: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 842: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 245: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 951: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 981: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 698: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 296: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 536: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 353: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 708: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 207: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 201: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 196: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 816: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 500: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 182: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 726: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 862: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 50: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 844: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 324: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 417: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 599: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 871: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 435: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 120: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 831: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 866: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 501: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 129: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 51: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 868: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 609: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 474: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 52: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 408: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 896: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 772: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 134: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 921: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 123: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 894: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 731: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 198: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 820: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 664: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 906: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 685: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 575: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 940: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 516: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 540: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 467: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 591: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 973: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 507: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 745: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 691: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 475: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 31: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 687: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 36: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 88: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 979: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 546: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 617: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 803: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 387: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 550: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 176: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 79: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 144: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 208: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 72: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*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.