#cost-performance
Research tagged cost-performance: explanations, evidence, limitations and connections between papers.
17 results
Choosing fewer, better tokens improves distillation
DIAL-OPD ranks response tokens by disagreement and probability, improving mean accuracy over full-token distillation in math tests.
DiffuPlex lets spoken-dialog models predict ahead and revise drafts
DiffuPlex predicts future conversation frames to reduce model calls while preserving tested interaction behavior.
Diffusion models can reveal memorized images before copies
A diffusion-model method probes hidden training-image memorization, with evidence bounded to tested datasets and one released model.
Byte models miss results that need calculation
A byte-level model’s loss-based patch signal finds computed results that entropy misses on arithmetic text.
A cache compressor rehearses likely answers before questions arrive
Attic-KV rehearses likely quoted answers and context anchors to improve cache compression before future questions are known.
Antenna agreement helps recover breathing under motion interference
A spatial adaptation method improves breathing-waveform recovery under simulated motion interference and lowers failures after real-data calibration.
A lattice method predicts better decoding schedules before generation
A lattice method estimates lost position dependence to rank decoding schedules before generation, with tested image evidence.
TokenRouter helps models hand off requests without restarting work
TokenRouter coordinates model handoffs and raised output-token rates over existing systems in tested workloads.
Reusing a student's first answers can improve distillation
Semi-on-policy distillation reuses a student's starting answers and outperformed standard distillation on 14 of 17 tested pairs.
Branching diffusion cuts the work of long text generation
Branching Latent Diffusion cuts computation for long text generation, while quality and single-sequence speed remain bounded.
SoftmaxLift reduces key retrieval for long-context attention
SoftmaxLift augments keys and queries to estimate attention with one exact search index and O(1/ε²) retrieved keys.
An optimizer change slows forgetting during adaptation
A training-time optimizer adjustment reduced old-domain validation-loss increases in seven stable Korean adaptation settings.
A direct formula makes private-training noise calibration less conservative
A direct noise-calibration method bounds membership tests in random-allocation DP-SGD and computes an exact reference.
COPC corrects stale weights and learning signals in language-model training
COPC corrects stale learning signals, with higher math and search accuracy than AReaL in two reported settings.
LayerRoPE turns layer depth into a signal for Transformer blocks
LayerRoPE generates Transformer normalization gains from shared, depth-adjusted values and had the lowest C4 validation loss at tested scales.
Jailbreak patches can also suppress benign prompts
A local evaluation finds jailbreak patches can correct nearby harmful prompts while also suppressing benign ones.
Model weight spectra rank robustness within model groups
A weight-based method ranks out-of-domain robustness within architecture and training groups, with limits across model families.