Jerry Liu

“Posted about optimizing OCRia-1 model speed on B200 GPU through speculative decoding and quantization.”

“Provided critical analysis of Cohere Parse 5, noting trade-offs between cost-competitiveness and performance in visual grounding, chart parsing, and fine-grained extraction.”

“Author discussing task-specific model routing as the best approach rather than one-size-fits-all routing strategies.”

“Announced LiteParse v2, rewritten in Rust for up to 100x speedups in PDF parsing and WASM edge deployment.”

“Published 'Files Are All You Need' discussing agent filesystem access instead of hundreds of tools.”