Search results

GRPO训练的PPO_KL和PG_clipfrac曲线和verl同样数据同样设置下差别很大，最终收敛结果也和verl训练差距很大

Status: Open.

#426 In modelscope/Trinity-RFT;

· chongzi1990 opened on Dec 3, 2025

mix chord算法复现问题

Status: Open.

#425 In modelscope/Trinity-RFT;

· 23557-l opened on Dec 3, 2025

SFT数据格式

Status: Open.

#418 In modelscope/Trinity-RFT;

· waltonfuture opened on Dec 1, 2025

关于cluster.node_num和cluster.gpu_per_node的疑惑

Status: Open.

#406 In modelscope/Trinity-RFT;

· tingjun-cs opened on Nov 24, 2025

关于 buffer.batch_size, buffer.train_batch_size, algorithm.repeat_times, runner_per_model, sync_interval等参数的疑惑

Status: Open.

#405 In modelscope/Trinity-RFT;

· tingjun-cs opened on Nov 24, 2025

[BUG] Timeout when running task

bugSomething isn't workingSomething isn't working

Status: Open.

#403 In modelscope/Trinity-RFT;

· Alital opened on Nov 22, 2025

请问如何断点续训？

bugSomething isn't workingSomething isn't working

Status: Open.

#399 In modelscope/Trinity-RFT;

· waltonfuture opened on Nov 21, 2025

mix-chord训练步数计算

Status: Open.

#396 In modelscope/Trinity-RFT;

· Alital opened on Nov 20, 2025

支持yarn扩展rollout模型上下文长度吗

Status: Open.

#389 In modelscope/Trinity-RFT;

· tingjun-cs opened on Nov 18, 2025

能否支持在 Experience 缓冲区中存储 Policy 末端隐含层输出

Status: Open.

#383 In modelscope/Trinity-RFT;

· SYSUzhouting opened on Nov 13, 2025

训练问题

Status: Open.

#365 In modelscope/Trinity-RFT;

· JiangWu0826 opened on Nov 5, 2025

请教一下Qwen2.5-3B-instruct用mix_chord训练，reward很低的原因

Status: Open.

#360 In modelscope/Trinity-RFT;

· caiyutong-eng opened on Nov 1, 2025