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 97: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 141: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 337: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 838: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 846: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 109: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 115: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 42: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 943: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 618: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 444: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 219: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 126: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 159: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 16: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 872: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 425: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 549: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 733: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 20: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 59: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 322: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 692: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 299: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 440: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 419: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 447: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 699: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 581: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 213: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 990: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 583: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 203: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 93: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 205: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 792: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 883: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 270: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 106: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 133: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 643: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 807: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 999: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 4: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 394: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 519: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 717: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 980: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 710: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 619: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 94: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 330: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 343: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 577: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 302: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 985: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 647: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 58: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 895: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 783: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 602: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 694: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 789: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 927: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 179: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 865: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 565: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 241: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 396: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 913: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 452: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 246: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 613: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 960: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 839: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 204: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 769: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 567: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 814: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 890: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 821: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 389: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 864: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 288: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 556: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 709: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 828: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 259: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 342: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 835: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 239: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 240: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 551: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 671: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 193: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 274: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 991: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 822: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 67: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*prec/8, saving = 1-comp/orig. Measured: 1000 tokens orig 6.1MB -> comp 338KB 94% saved target -80% PASS, full pipeli... | agentic | science | hard |
science | Science Hard 690: As researcher agent, explain PG-KVC Phase-Gated KV Cache 94% saving and PEP Phase-Entropy Predictor 100% accuracy for IKNN-Rl1-A1. Why phase-gated based on entropy? Formula for compression ratio? | Science: PG-KVC: KV cache entry precision gated by entropy H(p) = -sum P log P. If H < tau_low 0.5 (low entropy, predictable), use 1-bit (high compression), else 2-bit. Orig bytes = (k+v)*4, comp = (k+v)*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.