Search results

[Bug] AssertionError in fill_routing_replay when enabling MTP training for GLM-4.7

Status: Open.

#1556 In THUDM/slime;

· liujiahua123123 opened on Feb 6, 2026

MoE training: is there an API/hook to monitor expert activations?

Status: Open.

#1555 In THUDM/slime;

· tiesanguaixia opened on Feb 6, 2026

全异步返回数据是不是有bug? RolloutFnTrainOutput 格式？

Status: Open.

#1553 In THUDM/slime;

· HanelYuki opened on Feb 6, 2026

[feature request] Support multiple rollout worker for CPU-heavy custom <code>generate</code> function

Status: Open.

#1552 In THUDM/slime;

· leoleoasd opened on Feb 5, 2026

Speculative Decoding Standalone mode load draft model

Status: Open.

#1548 In THUDM/slime;

· ruiqiRichard opened on Feb 5, 2026

[BUG] LR scheduler double-counted when resuming from checkpoint

Status: Open.

#1546 In THUDM/slime;

· Surya-Gunukula opened on Feb 5, 2026

Do you have plans to support MiMo-V2-Flash?

Status: Open.

#1544 In THUDM/slime;

· Eisenhower opened on Feb 4, 2026

【模型支持】关于对Qwen3-Coder-Next模型的支持

Status: Open.

#1543 In THUDM/slime;

· Mrkkew opened on Feb 4, 2026

run-qwen3-30B-A3B-int4.sh megatron在加载模型时超时

Status: Open.

#1540 In THUDM/slime;

· 191220042 opened on Feb 3, 2026

[Usage] does update_weights_from_tensor save weights data ?

Status: Open.

#1536 In THUDM/slime;

· xgwang opened on Feb 3, 2026

Is glm-4.7-flash MTP training in RL stage supported for now?

Status: Open.

#1534 In THUDM/slime;

· ifififa opened on Feb 2, 2026

Policy Collapse and Gibberish Rollouts in Search-R1 Early Training when TIS is Enabled (3B Model)

Status: Open.

#1533 In THUDM/slime;

· Teglacer opened on Feb 2, 2026