Search results

Error: Qwen3.5-35B-A3B lora sft with mcore-adapter, error occurs when saving a ckpt.

Status: Open.

#411 In alibaba/ROLL;

· Unofish opened on Mar 30, 2026

Feature Request: Trajectory-level formulation mode for AgentNativeStepEnvManager

Status: Open.

#409 In alibaba/ROLL;

· shamanez opened on Mar 28, 2026

expandable_segments:True hardcoded in CUDA platform files causes error on older kernel; PYTORCH_CUDA_ALLOC_CONF ignored

Status: Open.

#408 In alibaba/ROLL;

· cchen-hhz opened on Mar 28, 2026

LR scheduler exhausts early in agentic training with AgentNativeStepEnvManager

Status: Open.

#407 In alibaba/ROLL;

· shamanez opened on Mar 27, 2026

[Feature Request] Add official Qwen3.5 Sokoban agentic example

Status: Open.

#406 In alibaba/ROLL;

· yingyibiao opened on Mar 27, 2026

Add OpenReward Intergration

Status: Open.

#405 In alibaba/ROLL;

· shamanez opened on Mar 27, 2026

Add Trackio as a New Backend for Experiment Visualization

Status: Open.

#402 In alibaba/ROLL;

· ParagEkbote opened on Mar 25, 2026

Torch 2.8.0 + vLLM 0.10.2 开启 use_kl_loss 后训练qwen2.5-vl-3B-agentic崩溃，Torch 2.6.0 + vLLM 0.8.4 正常

Status: Open.

#398 In alibaba/ROLL;

· SF-evil opened on Mar 24, 2026

Very slow asynchronous GRPO training on 8×A100

Status: Open.

#394 In alibaba/ROLL;

· 5SSjw opened on Mar 22, 2026

[Type Hints] Add type annotations to utility functions in roll/utils/functionals.py

Status: Open.

#387 In alibaba/ROLL;

· guozhihao-224 opened on Mar 18, 2026

Packing support in mcore-adapter VL models

Status: Open.

#386 In alibaba/ROLL;

· JCode012 opened on Mar 18, 2026

how to enable torch.compile when use Megatron-LM as training backend?

Status: Open.

#381 In alibaba/ROLL;

· robotsp opened on Mar 16, 2026