KDA CUDA Trainer

A compact, trainable causal language model example built with Kimi Delta Attention (KDA), including CUDA (FLA kernel) and PyTorch reference paths.

Quick Start

Train with built-in tiny text corpus:

python -m kda_train.train --device cuda --steps 1000

Repository

GitHub README

Official Website Traffic / 官网访问统计

Last 30 days: page views, visits, and top countries / regions.

Website statistics for the last 30 days: page views, visits, and country / region distribution.

Source: Cloudflare Web Analytics. Visits are not unique visitors (UV). These figures cover this website; GitHub repository views are excluded. / 数据为本站近 30 天浏览量、访问次数及国家/地区分布，访问次数不等于去重访客数。

GitHub README traffic is tracked separately with VisitorBadge and FlagCounter. View repository counters / 查看仓库访问统计.