ILHO AHN

Mini Research

Experiments driven by personal curiosity

LLM ALIGNMENT Alignment Data Map: Timing of SimPO Boundary Crossings and Model-Specific Differences Models trained on ADM HighAvg or Random data are compared on the same held-out preference pairs, tracing SimPO boundary crossings across model variants and checkpoints.

LLM EVAL Fixed-Budget Tool-Use DPO: Call Correctness and Follow-Up Decisions Under matched training budgets, structure-focused DPO led call correctness and decision-focused DPO led behavior selection. Repeated recipe rankings are examined alongside correct and incorrect follow-ups.

1. LLM ALIGNMENT Alignment Data Map: From Measurements to Preference-Pair Supervision This note traces how Alignment Data Map coordinates vary with the reference answer and text processing, and how selected instructions become preference pairs used for training.
2. LLM TOOL USE DiaTool-DPO Reconstruction: Preference Accuracy and Tool-Use Behavior A DiaTool-DPO reconstruction connecting preference ranking, repeated training responses, and changes in missing-field handling and complete-call success.
3. LLM SYSTEMS DPO Preference Packing: Dense Masks and Sparse Execution A DPO systems note separating shared-prompt layout requirements from dense and sparse execution, with recorded Qwen3-8B training-cost comparisons.
4. BIO ML Carbon-3B: Measuring 6-mer Token Phase Sensitivity A test of how much 6-mer token phase changes Carbon-3B scores relative to score IQR across 500 BRCA2 MAVE SNVs, and how the corresponding FNS pipeline differs.
5. BIO ML Auditing Structural-Signal Interpretation in OpenBind Prediction Scores A reproducible benchmark audit comparing OpenBind prediction-score correlations with property baselines and ligand-only controls to examine the limits of structural-signal interpretation.
6. PROTEIN ML Low-label Protein Fitness 실험: Frozen ESM2와 LoRA의 label budget 비교 라벨을 많이 모으기 어려운 protein fitness 예측에서 frozen ESM2 embedding만 쓰는 방법과 LoRA로 조금 적응시키는 방법을 같은 label budget 안에서 비교한다.
7. LLM EVAL Gemma3 4B와 Gemma4 E4B의 한국어 SFT 비교 한국어 holdout에서 Gemma3 4B와 Gemma4 E4B의 추가 SFT, 답안 구성, 입력 노이즈를 비교했다. Judge 선택률·형식 준수와 표현 중복, 절대 점수·하락폭을 구분해 해석한다.