PokeLu

梯子下更新RSS订阅、tracker list失败 RSS and tracker list updating failed using proxy

26

### qBittorrent & operating system versions qBittorent: v4.6.3.10 x64 Operating System: Windows Server 2022 Datacenter Evaluation 21H2, 2023/‎11/‎11, 20348.2322 Qt: 6.4.3 Libtorrent: 1.2.19.0 Boost: 1.83.0 OpenSSL: 1.1.1w zlib: 1.3 ###...

What is the correct way to evaluate thinking model via mmlu_generative?

1

As stated in the title, I was trying to evaluate the generative mode mmlu for thinking model such as: ``` # export HF_ENDPOINT=https://hf-mirror.com; # export HF_DATASETS_CACHE='/root/.cache' export LOGLEVEL=DEBUG; export VLLM_WORKER_MULTIPROC_METHOD=spawn;...

``` OSError: [Errno 36] File name too long: '/lm-evaluation-harness/.process_cache/requests-mmlu_elementary_mathematics_generative-5shot-rank0-world_size1-chat_template-fewshot_as_multiturn-tokenizer__ckpts__Qwen3-4B-Base__sft__AM-DeepSeek-R1-0528-Distilled-Nothinking__ep2.57669194e25afae85ea14f096de3ea7c6247d45e62770904575c34cbce2478f0.pickle' ``` As aforementioned, I encountered a `file name too long` bug for the full mmlu-generative task while using vllm to deploy...

Fully Async Recipe OOM with only 8192 response length

6

Hello, thanks for the developers for finally get the fully async policy working in Verl. In my personal trial, I modified [the geo3k_qwen25vl_7b_megatron_4_4 script](https://github.com/volcengine/verl/blob/main/recipe/fully_async_policy/shell/geo3k_qwen25vl_7b_megatron_4_4.sh) to test gsm8k on Qwen3 4B...

是否支持一键同步库内所有番剧的观看进度？

2

第一次使用[jellyfin-plugin-bangumi](https://github.com/kookxiang/jellyfin-plugin-bangumi)，请问是否支持将库中所有番的观看进度一键同步至bangumi？另外如果番剧未在bangumi手动收藏则无法同步：`[ERR] [64] Jellyfin.Plugin.Bangumi.PlaybackScrobbler: report playback status failed: "System.Net.Http.HttpIOException: Bad Request: you need to add subject to your collection first (InvalidResponse)`。当前我使用autobangumi/anirss构建影音库，每次新番入库是否仍需要手动收藏？