Benchmarking Korean Language Models: CLIcK, KMMLU, and Cultural Nuance
A source audit evaluating Korean LLM benchmark frameworks including CLIcK, KMMLU, and HAERAE, assessing cultural nuance, linguistic grounding, and reasoning.
Fodev JEO
LATEST INVESTIGATIONS
Source audits & field notes
Foundational AI Architectures: Mapping the Canonical Papers of Modern Deep Learning
A source audit analyzing the canonical architectures across transformers, vision encoders, multimodal CLIP, diffusion, PEFT, and attention kernels.
Vector Embeddings Architecture: Contrastive Learning, MTEB, and Matryoshka Projections
A source audit examining vector embedding architectures, contrastive learning objectives, MTEB evaluation benchmarks, and Matryoshka dimension truncation.
Implementing Llama 3 in Pure NumPy: First-Principles Tensor Inference
A source audit analyzing first-principles Llama 3 inference in pure NumPy, breaking down RoPE, SwiGLU, RMSNorm, and Grouped-Query Attention mechanics.
Building Systematic RAG Workflows: Beyond the Naive Vector Search Pattern
A systems engineering guide to production RAG workflows: contextual routing, dynamic chunking, hybrid retrieval, and automated evaluation harnesses.
Elasticsearch Architecture: Inverted Indices, Observability, and Hybrid Search
A source audit analyzing Elasticsearch architecture, evaluating inverted indices, real-time log analytics, SIEM telemetry, and 2026 hybrid vector search.
Vision-Language Architectures: Cross-Attention vs Autoregressive Fusion
A source audit analyzing Vision-Language Models (VLMs), comparing cross-attention versus fully autoregressive token fusion, dynamic aspect ratios, and sub-image tiling.
MMLU-Pro Evaluation: Rigorous Multi-Task Reasoning and 10-Option Search Spaces
A source audit analyzing TIGER-Lab’s MMLU-Pro benchmark, evaluating 10-option search spaces, chain-of-thought dependencies, and model reasoning degradation.
LLM Evaluation Frameworks: MMLU-Pro, Length-Controlled AlpacaEval, and Arena Elo
A source audit analyzing LLM evaluation methodology, evaluating MMLU-Pro reasoning, length-controlled AlpacaEval 2.0, LMSYS Arena Elo, and 2026 benchmarks.
Online Iterative DPO Architecture: Eliminating Offline Alignment Drift
A source audit analyzing Salesforce’s Online Iterative DPO framework, examining proxy reward models, off-policy drift, and synthetic alignment feedback loops.
인기 태그
Search article titles, categories, and tags. Full text is searched when needed.
Type to search articles.