01▲TeleZK-FL: enabling trustless and verifiable remote patient monitoring via quantized zero-knowledge federated learning frontiersin.org Millisecond ZK proofs on Raspberry Pi telehealth clients, as long as you’re fine with KZG and a bit less AUC.blockchainedge-computingfederated-learninghealthcaremedical-aiprivacyquantizationsnarktelehealthtwitterzk0 pts/veramarsh/1 day ago/1 comment
02▲Smaller, faster, safer: running Kimi and GLM at scale blog.cloudflare.com Cloudflare's twist on serving big open models: FP8 KV caches, INT4 weights, cache checks, and no measured accuracy loss.blogsgpuinferencelatencyllmmemoryquantizationsafetyservingthroughput0 pts/nonce12/10 days ago/1 comment
03▲Fine-Tuning from First Principles: LoRA, QLoRA, and Serverless Fine-Tuning on Crusoe debnsuma.github.io Full fine-tuning is the expensive part, this derives LoRA/QLoRA from scratch and ends with a serverless PII redactor.deep-learningfhefine-tuninglinkedinllmllm-inferencelorampcpqcpytorchqloraquantizationserverlesssnarkzk0 pts/mara/13 days ago/discuss