Independent research on inference optimization, constitutional AI architectures, and empirical safety evaluation. Each paper is backed by reproducible technical reports and artifact-level provenance from a 752,000+ measurement program.
Author: Sahil Kadadekar · Independent research
5 papers submitted with PDFs, artifact manifests, and venue checklists complete. Now under peer review.
Independent upstream bugs in PyTorch and Triton jointly produce the torch.compile decode crash. Triton minor-version ablation on the same GPU flips the conclusion. Benchmark identity is a 5-tuple (GPU, Triton, PyTorch, cache, compile mode). Companion to upstream PR #175562.
Simpson's paradox in the safety-quality relationship. Refusal Template Stability Index calibrated with LOOCV across 51 model-format cells (GGUF + AWQ + GPTQ), inter-judge κ = 0.873.
Q2_K is the recurring vulnerability threshold for many-shot and long-context attacks. Message-array vs faux-dialogue prompt formatting (92% vs 0% ASR) across 4 model families. Format mediates effect more strongly than quantization alone.
16,783 samples across production-scale 70B target + 8B draft pairs (adversarial draft, quantized draft, non-greedy decoding). Zero measurable safety degradation, contradicting the SSD premise. Strong null result.
8 attack strategies × 4 models × 6 quantization levels: 10,600 conversations, 37,825 judge labels. Threshold-specific shift in risk rather than universal multi-turn amplification.
Synthesis papers and methodology work derived from the published technical report archive.
Phase 1 safety flips at ~0.58% vs capability ~0.14% under controlled batching. Refusal-to-compliance dominant direction. Reduced true-batching validation reaches ~99.4% agreement with synchronized dispatch.
Synthesis paper. Quantization drives 57% of total safety cost, backend choice 41%, concurrency 2%. Chat template divergence can induce larger safety shifts than numerical precision.
Capacity planning as a fitted systems problem. Backend choice, context length, and memory pressure all materially change the feasible operating regime. Planner quality should be judged by validation against explicit targets, not analytic elegance.
Recasts "which language wins" as "which system design preserves throughput." Python and Rust near-parity on throughput; architecture and concurrency strategy drive larger differences. Dual Ollama achieves 99.4% multi-agent efficiency.
LLM serving-stack differences are mechanistic, not benchmark trivia. Continuous batching drives throughput scaling with agent count via 77-80% kernel reduction. Backends differ in effective serial fraction. 2.25× throughput gain at N=8 explained mechanistically.
KV-cache quantization is a serving-layer perturbation that touches retained attention state. 5-phase paired study on FP16 vs FP8 across 24K records, 3 models. Headline result is a null: no Holm-significant safety effect detectable at α=0.05, 80% power. Operational rule: workload-specific paired eval, not pre-approval.