v1.5.8 Beta48 Release

v1.5.8 Release Latest

This is an experimental project currently in beta, and there are still tons of bugs.
这是一个实验性的项目，目前仍处于 beta 阶段，依然有成吨的问题。
Please check back often for updates.
请常回来看看有没有更新。

Highlight / 更新亮点

• Fix NEWBIE startup rejection when use_cache=false or newbie_two_phase_execution=false, and switch these modes to compatibility execution instead
• 修复 NEWBIE 启动拦截问题：在 use_cache=false 等配置下启动被拦截的问题，现已支持自动切换至兼容模式。

• Add transient cache cleanup for NEWBIE compatibility runs so temporary cache files are removed automatically after the run finishes
• 改进 NEWBIE 临时缓存管理：增强了对损坏或残缺 safetensors 缓存文件的校验；兼容模式下的临时缓存现在会在运行结束后自动清理。

• Add automatic full failure report output for NEWBIE cache failures while keeping console output readable
• 改进 NEWBIE 缓存报错输出：在保留控制台简洁输出的同时，缓存失败时会自动写出完整失败报告，便于排查问题。

• Tighten NEWBIE resolution and bucket validation, including explicit multiple-of-8 checks and invalid bucket range rejection
• 补强 NEWBIE 配置校验：新增分辨率与 Bucket 参数的 8 倍数强制检查，并拦截非法配置。

• Improve NEWBIE cache integrity checks so invalid or incomplete safetensors cache files are no longer treated as valid cache
• 提升 NEWBIE 缓存完整性判断：损坏或缺少关键字段的 safetensors 缓存文件不再被误判为有效缓存。

• Align lulynx_experimental_core_enabled runtime default behavior with the UI schema default
• 修复 lulynx_experimental_core_enabled 默认值问题：同步了 lulynx_experimental_core_enabled 在运行时与 UI Schema 的默认值，解决配置漂移问题。

• Fix ANIMA train_llm_adapter parsing so both boolean and string inputs work correctly without crashing
• 修复 ANIMA 的 train_llm_adapter 参数解析：修复了 train_llm_adapter 对布尔值与字符串的解析逻辑，避免类型冲突导致崩溃。

• Protect ANIMA cache_latents execution with guaranteed VAE cleanup, ensuring the VAE returns to CPU and device memory is cleaned even if caching fails midway
• 修复 ANIMA cache_latents 异常清理问题：完善了 cache_latents 的异常清理机制，即使缓存中途失败也会强制回收 VAE 显存，防止状态异常。

• Limit ANIMA padding mask cache growth to prevent unbounded memory expansion during long or mixed-resolution training sessions
• 修复 ANIMA padding mask 缓存无限增长问题：为 Padding Mask 缓存增加了上限，解决了长时或混合分辨率训练中的内存膨胀风险。