lizhanyuan 252d2f79ce fix(eval): 修复vllm_eval截图排序bug并对齐reeval逻辑 ...

- 修复_load_screenshots_from_dir中截图按字符串排序导致step_9被误判为最终帧的bug，改为数字排序
- 对齐reeval.py的prompt逻辑：明确要求模型优先检查最终截图（STEP 1 EXAMINE FINAL SCREENSHOT FIRST）
- 评估temperature从0.7降至0.2提升一致性
- 新增batch_reeval.py：基于test_final.json批量重评测已有轨迹
- 新增reeval.py：单任务重评测脚本（final-frame-anchored evaluation）
- test_final.json新增avogadro(11题)和origin(8题)

2026-03-27 14:34:32 +08:00

..

controllers

Add Claude Sonnet 4.5 support and improve action handling (#362)

2025-11-14 13:54:32 +08:00

evaluators

fix(eval): 修复vllm_eval截图排序bug并对齐reeval逻辑

2026-03-27 14:34:32 +08:00

providers

feat: 新增 Proxmox provider 支持及 inject_steps 参数

2026-03-04 16:39:08 +08:00

server

feat(server): add cross-platform support and improve screenshot handling

2026-01-30 16:27:49 +08:00

__init__.py

Refactor examples; Start to load examples into benchmark; vlc initialization

2023-12-25 00:24:13 +08:00

actions.py

Refactoring VMware Integration and Implementing AWS Support (#44)

2024-06-15 20:52:29 +08:00

desktop_env_os_symphony.py

fix(os_symphony_evaluation) (#410)

2026-01-04 15:56:51 +08:00

desktop_env.py

feat: 新增 Proxmox provider 支持及 inject_steps 参数

2026-03-04 16:39:08 +08:00