ann ar aws block-wise-scaling bpe cache cfg chatgpt compositions conf cpu crdt db design diffusion dp dual-ar encodec firefly-gan flow fsq gcp gemma geo geohash gfsq git golang gpu hashmap hierarchical-navigable-small-world hnsw hugo idgen inference-optimization integer-parsing inverted-file-index inverted-index knn kv langchain linux llama llm locality-sensitive-hashing lru lsh macbook mel-spectrogram memory-profiling ml model mp multimoding nar nlp nn non-g2p octopress oneday openai openvoice parallelism performance-analysis ping pool product-quantization rag random-projections rapids-libcudf rapids-raft redis rust simd similarity-search streaming string supervised-semantic-speech-tokenizer tf tools training transformer transformers translate vall-e vits voice vq-gan wal 全栈 工程师 方案 旅游 直播 网络编程 随笔记录