v1.1.0

embed dependencies

v1.0.9

Features

use model state cache to achieve 5x - 50x faster preparation time for generation
support for rwkv-4-world
preload preset_system
improve api concurrency performance
experimental macOS/Linux support

Fixes

the configs page now always displays the currently selected non-local model so that other models can be selected properly
sha256 check for model deduplication

chore

add FAQs to About

v1.0.8

fix startup status detect
update defaultModelConfigs
update rwkv_generate
update completion presets
update home page
add novel models source

v1.0.7

fix cuda40 kernel

v1.0.6

improve upgrade process

v1.0.5

Completion Page
fix api completion_lock
fix manifest programFiles version to enhance cdn real-time performance
add role: "system" support
support nvidia 16xx custom cuda kernel
update nvidia 40xx custom cuda kernel
improve upgrade process
add api host setting

v1.0.4

improve interaction and avoid user mistakes

v1.0.3

update manifest
update defaultconfigs
fix duplicate switchModel()

v1.0.2

Custom CUDA kernel acceleration

v1.0.1

fix completion format
add extra updates source
add compatible /v1/completions API
improve compatible API