Search results

mask utils在glm系列会少算stop token

Status: Open.

#1490 In THUDM/slime;

· Fu-Dayuan opened on Jan 25, 2026

Training hangs indefinitely after rollout phase completes on 8 B200 GPUs with CP=4, TP=2

Status: Open.

#1487 In THUDM/slime;

· euclidgame opened on Jan 23, 2026

Slime + Megatron backend: training frequently hangs after SGLang CUDA graph capture & server startup (Uvicorn ready), no further progress

Status: Open.

#1484 In THUDM/slime;

· tiesanguaixia opened on Jan 23, 2026

About GLM-4.7-Flash

Status: Open.

#1482 In THUDM/slime;

· YukinoshitaKaren opened on Jan 23, 2026

loss mask question

Status: Open.

#1477 In THUDM/slime;

· glennccc opened on Jan 21, 2026

请求支持 rollout 粒度的 sglang 请求路由提高多轮 rollout 的吞吐量

Status: Open.

#1469 In THUDM/slime;

· zchuz opened on Jan 20, 2026

Do not support Gemma3 in convert_hf_to_torch_dist

Status: Open.

#1468 In THUDM/slime;

· Zhaoyi-Yan opened on Jan 20, 2026

cannot import name 'GPU_MEMORY_TYPE_CUDA_GRAPH' from 'sglang.srt.constants'

Status: Open.

#1462 In THUDM/slime;

· CHRIS123540 opened on Jan 20, 2026

The Sglang version in docker image

Status: Open.

#1455 In THUDM/slime;

· AIR-hl opened on Jan 19, 2026

Bug: the generate_rollout in sft_rollout.py has not truncation

Status: Open.

#1450 In THUDM/slime;

· MikaStars39 opened on Jan 18, 2026

on_policy_distillation, advantage_estimator

Status: Open.

#1449 In THUDM/slime;

· EthanChen1234 opened on Jan 18, 2026

An error occurred while converting Llama3.2-1B-Instruct to Megatron ckpt.

Status: Open.

Bug

#1430 In THUDM/slime;

· SmileShaun opened on Jan 16, 2026