Llama 3 70B on AWS Inferentia2: Optimum Neuron, NeuronCore-v2, and TGI Serving
A source audit analyzing Llama 3 70B inference on AWS Inferentia2 using Optimum Neuron, NeuronCore-v2 compilation graphs, and Hugging Face TGI serving.
Fodev JEO
LATEST INVESTIGATIONS
Source audits & field notes
The Open LLM Curriculum: Deconstructing Scientist and Engineer Competency Maps
A source audit analyzing Maxime Labonne’s LLM Course, mapping core engineering competencies across pretraining mathematics, post-training, and serving.
ViViD: Video Virtual Try-On Using Diffusion Models and Temporal Attention
A technical source audit of ViViD (arXiv:2405.11794), evaluating diffusion-based video virtual try-on, garment spatial-temporal attention, and cross-frame fidelity.
Hugging Face and Microsoft Alliance: AMD MI300X, ONNX Runtime, and WebGPU
A source audit examining the Microsoft-Hugging Face partnership, analyzing AMD Instinct MI300X ROCm acceleration, ONNX Runtime optimization, and WebGPU.
How Operating Systems Execute Code: Virtual Memory, Dynamic Linking, and CPU Pipelines
A source audit analyzing OS process execution, ELF/Mach-O loaders, virtual memory page tables, dynamic linker relocations, and CPU out-of-order pipelines.
Generative AI Taxonomy: Core Mechanisms from Transformers to Flow Matching
A source audit establishing the foundational taxonomy of Generative AI: autoregressive sequence modeling, diffusion SDEs, flow matching, and latent topologies.
Unsloth Kernel Architecture: Custom Triton Backpropagation and QLoRA Memory Scaling
A source audit analyzing Unsloth’s custom Triton kernels, manual backpropagation derivation, RoPE fusion, and QLoRA memory reductions in Hugging Face pipelines.
Ten Foundational Algorithms That Shaped Distributed Systems and AI
A technical audit analyzing the foundational algorithms powering modern computing, from Cooley-Tukey FFT and Dijkstra to PageRank and Transformer attention.
Graph Analytics and GNNs: From Network Topology to Modern GraphRAG
A source audit analyzing graph representation learning, message-passing neural networks, topological community detection, and modern GraphRAG architectures.
Google I/O 2024 vs GPT-4o: The Architectural Battle for Native Multimodality
A source audit comparing Google I/O 2024 and OpenAI’s GPT-4o, examining native omni-modal tokenization, 2M context windows, and real-time speech synthesis.
인기 태그
Search article titles, categories, and tags. Full text is searched when needed.
Type to search articles.