1 Release published by 1 person

• v3.5.2 Patch release v3.5.2

  published Jun 20, 2025

29 Pull requests merged by 6 people

• compat transformers==4.52 (vlm)

  #4738 merged Jun 26, 2025

• [grpo] check liger & sp

  #4734 merged Jun 26, 2025

• [grpo] fix max_step for dataloader when applying sequence parallel

  #4731 merged Jun 26, 2025

• [quant] Support fp8

  #4729 merged Jun 26, 2025

• support Kimi-VL-A3B-Thinking-2506 & Kimi-Dev-72B

  #4719 merged Jun 25, 2025

• [doc] simplify environment variables & update best practices documentation

  #4715 merged Jun 25, 2025

• [grpo] fix colocate seed

  #4712 merged Jun 25, 2025

• [megatron] support rednote-hilab/dots.llm1.inst

  #4707 merged Jun 25, 2025

• [megatron] support DeepseekV2ForCausalLM and DeepseekV3ForCausalLM

  #4659 merged Jun 25, 2025

• fix links

  #4690 merged Jun 24, 2025

• [feat] support fine-tuning of reranker models

  #4671 merged Jun 24, 2025

• [grpo] fix grpo pt

  #4683 merged Jun 24, 2025

• [rollout] fix dp args

  #4678 merged Jun 23, 2025

• [doc] fix doc

  #4675 merged Jun 23, 2025

• [doc] fix image link

  #4674 merged Jun 23, 2025

• docs: correct typo "resonse" to "response"

  #4672 merged Jun 23, 2025

• [channel loss]support packing & padding free

  #4666 merged Jun 23, 2025

• [docs] update docs

  #4665 merged Jun 23, 2025

• [dataset] fix grounding_dataset

  #4664 merged Jun 23, 2025

• [grpo] refactor multi turn & support async engine & refactor grpo docs

  #4380 merged Jun 23, 2025

• [template] optimize remove_unused_columns

  #4661 merged Jun 22, 2025

• [gkd] support use_logits_to_keep/padding_free/packing & update gkd shell

  #4658 merged Jun 21, 2025

• [docs] update gkd

  #4657 merged Jun 20, 2025

• compat megatron-core 0.11

  #4655 merged Jun 20, 2025

• [megatron] fix eval data_collator

  #4654 merged Jun 20, 2025

• fix device_map & ddp rank0

  #4650 merged Jun 20, 2025

• fix packing & load_from_cache_file

  #4649 merged Jun 20, 2025

• [model] fix model_meta

  #4647 merged Jun 20, 2025

• [template] optimize get_length

  #4641 merged Jun 20, 2025

4 Pull requests opened by 4 people

• solve the default 'template_backend' bug in llm.tempalte.base.Templte._encode

  #4669 opened Jun 23, 2025

• Refactor Web-UI

  #4687 opened Jun 24, 2025

• [grpo]Tool rl: add reward func for ToolRL

  #4694 opened Jun 24, 2025

• [megatron] support fp8

  #4730 opened Jun 26, 2025

124 Issues closed by 10 people

• GRPO怎么自定义format reward

  #4667 closed Jun 26, 2025

• [grpo] loading BERT model in reward

  #4580 closed Jun 26, 2025

• GRPO训练中Loss和grad_norm一直为0

  #4570 closed Jun 26, 2025

• GRPO什么时候支持多机megatron训练

  #4558 closed Jun 26, 2025

• GRPO训练reward的std始终为0

  #4512 closed Jun 26, 2025

• 多机训练使用--vllm_mode server 会卡死无法运行

  #4532 closed Jun 26, 2025

• GRPO Qwen3 32B training torch issue

  #4491 closed Jun 26, 2025

• qwen3强化训练，grpo训练结束后，爆通信错误

  #4170 closed Jun 26, 2025

• The expanded size of the tensor (8) must match the existing size (5) at non-singleton dimension 0.

  #4056 closed Jun 26, 2025

• 训练结束报错/data/chatglm/retrieval_agent_new/ms_swift_train/ms-swift/swift/cli/rlhf.py FAILED

  #4302 closed Jun 26, 2025

• dapo时在UserWarning: None of the inputs have requires_grad=True. Gradients will be None一直卡住，直至timeout

  #4050 closed Jun 26, 2025

• 用grpo训练qwen2.5-7b-instruct出现!!!!

  #4060 closed Jun 26, 2025

• 训练正常 eval时报assert error

  #4081 closed Jun 26, 2025

• Batch size in GRPO.

  #4341 closed Jun 26, 2025

• grpo训练奖励函数注册失败

  #4351 closed Jun 26, 2025

• GRPO数据传递失败

  #4362 closed Jun 26, 2025

• Qwen-Omni 全量微调grpo报错ValueError: `max_new_tokens` must be greater than 0, but is -16384

  #4392 closed Jun 26, 2025

• GRPO微调多模态训练报错

  #4470 closed Jun 26, 2025

• 双卡A6000使用GRPO微调Qwen2.5-VL-3B会OOM吗？

  #4477 closed Jun 26, 2025

• RTX3090上运行sft-rlhf-grpo微调，报错：torch.distributed.DistBackendError: [3] is setting up NCCL communicator and retrieving ncclUniqueId from [0] via c10d key-value store by key '0', but store->get('0') got error: wait timeout after 1800000ms,

  #3612 closed Jun 26, 2025

• Any plans to support megatron for GRPO training?

  #3760 closed Jun 26, 2025

• LLava 跑GRPO 无法跑通

  #3928 closed Jun 26, 2025

• QWQ：GRPO训练无法跑通，报错”RuntimeError: ACL stream synchronize failed, error code:107020“

  #3932 closed Jun 26, 2025

• While training GRPO, I noticed that my model crashes. Its loss is 0, its grad_norm and kl are both Nan, and it completes as “!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!””

  #3930 closed Jun 26, 2025

• GRPO训练中间一部分后报错

  #3771 closed Jun 26, 2025

• grpo训练卡住，一直显示一下问题。

  #3794 closed Jun 26, 2025

• GRPO训练报错

  #3769 closed Jun 26, 2025

• Various traceback error during GRPO training

  #3836 closed Jun 26, 2025

• 贡献一个dockerfile吧，这个测试了 多模态的grpo训练 可以基本可以复现示例里面的结果

  #3812 closed Jun 26, 2025

• GRPO 算法如果设置 reward_model 而不是--reward_funcs ，reward模型和 model都加载到一张卡里去了

  #3843 closed Jun 26, 2025

• Meet GPU OutOfMemory in GRPO training

  #3848 closed Jun 26, 2025

• grpo训练32b模型OOM

  #3871 closed Jun 26, 2025

• GRPO 训练100 steps后性能骤降，请问是什么原因

  #3876 closed Jun 26, 2025

• Bug! Checkpoint resume failure - deepspeed different DP size. Is there a quick checkpoint converter anywere?

  #3989 closed Jun 26, 2025

• Bug! Help! MS-SWIFT GRPO + LoRA training hung/stuck after training 1 step from full merged model merged from lora adapter

  #3990 closed Jun 26, 2025

• if sleep_level > 0, gradient_accumulation_steps will be forced to 1

  #3943 closed Jun 26, 2025

• The GRPO training process hangs for multi-node training.

  #3934 closed Jun 26, 2025

• NPU环境训练速度问题

  #3331 closed Jun 26, 2025

• 求一个能8卡A100使用GRPO跑通Qwen2.5 72B模型的脚本

  #3416 closed Jun 26, 2025

• GRPO 训练时使用2个节点并且设置--num_infer_workers 2 时会报错

  #3393 closed Jun 26, 2025

• 基于qwenvl-7b-instruct训练grpo，eval过程会oom

  #3541 closed Jun 26, 2025

• 4*v100环境执行lora_vllm脚本报错：Assertion `!(srcMmaLayout && dstMmaLayout && !srcMmaLayout.isAmpere()) && "mma -> mma layout conversion is only supported on Ampere"' failed.

  #3549 closed Jun 26, 2025

• 单机多卡跑grpo，多个step后会报错

  #3576 closed Jun 26, 2025

• Loss goes to 0, Gibberish Outputs

  #3582 closed Jun 26, 2025

• 日志怎么添加训练数据中的字段

  #3591 closed Jun 26, 2025

• 多机多卡GRPO assert self.cpu_group is not None

  #3583 closed Jun 26, 2025

• 设置NPROC_PER_NODE后会直接报错 failed (exitcode: -11) local_rank: 1

  #3611 closed Jun 26, 2025

• GRPO算法训练，后期训练时，显存暴增

  #3600 closed Jun 26, 2025

• grpo 固定seed，结果依旧不可复现

  #3607 closed Jun 26, 2025

• gemma3使用grpo用vllm的bug

  #3660 closed Jun 26, 2025

• 【bug】Failed to open local file in cache

  #3667 closed Jun 26, 2025

• [Bug]: RuntimeError: setup failed!

  #3662 closed Jun 26, 2025

• 使用GRPO训练llava-1.5以及qwen2-vl时，使用vllm推理，在eval时报错

  #3666 closed Jun 26, 2025

• 有没有4*V100能跑起来GRPO的训练脚本和环境配置呀？

  #3671 closed Jun 26, 2025

• ValueError: RLHF do not support sequence parallel

  #3673 closed Jun 26, 2025

• Hanging after tqdm starts [COLOCATE MODE]

  #3702 closed Jun 26, 2025

• GRPO max_grad_norm seems don't work

  #3713 closed Jun 26, 2025

• It is recommended to use a dedicated device for vLLM

  #3719 closed Jun 26, 2025

• npu环境GRPO训练，使用vllm时，官方脚本无法正常启动，其他脚本则可以

  #3726 closed Jun 26, 2025

• GRPO 训练，数据格式解析有bug

  #3728 closed Jun 26, 2025

• TypeError: embedding(): argument 'indices' (position 2) must be Tensor, not NoneType

  #3730 closed Jun 26, 2025

• Support Ulysses in Swift

  #3731 closed Jun 26, 2025

• GRPO tutorial bug: world_size (8) is not equal to tensor_model_parallel_size (4) x pipeline_model_parallel_size (1)

  #3739 closed Jun 26, 2025

• 多模态qwen2.5-vl-3B,grpo实验报错

  #3398 closed Jun 26, 2025

• grpo微调deepseek v2，训练过程中到eval阶段，就会卡住，然后就会停止训练

  #3528 closed Jun 26, 2025

• 请问如何在grpo中配置自定义的数据集路径，并进行数据格式转换？

  #3525 closed Jun 26, 2025

• 2workers_async_iterations2_vllm help

  #3522 closed Jun 26, 2025

• Bug in GRPO best practices document!

  #3501 closed Jun 26, 2025

• unhashable type: 'list'

  #3490 closed Jun 26, 2025

• 请求支持GRPO训练中，vllm推理后端支持多张卡🙏 Request for support for using multiple cards in the vLLM inference backend during GRPO training

  #3477 closed Jun 26, 2025

• 使用GRPO进行Qwen2.5-vl-7B-Instruct训练，报错：无法多卡训练，只能加载1张卡并oom

  #3404 closed Jun 26, 2025

• GRPO训练功能建议

  #3415 closed Jun 26, 2025

• GRPO 训练loss和reward异常

  #3372 closed Jun 26, 2025

• grpo 多机多卡训练timeout

  #3343 closed Jun 26, 2025

• GRPO训练LLAVA CUDA Error

  #3264 closed Jun 26, 2025

• GRPO LLava 训练报错，无法多卡训练，1卡可以

  #3228 closed Jun 26, 2025

• GRPO 4卡A100训练BUG

  #3223 closed Jun 26, 2025

• 如何对deepseek r1做sft和grpo微调

  #3211 closed Jun 26, 2025

• 使用GRPO 使用我已经训练的LLava模型加载问题

  #3195 closed Jun 26, 2025

• GRPO deepspeed lmdeploy训练InternVL2d5 报错

  #3151 closed Jun 26, 2025

• Using Unsloth in conjunction with GRPO to train a model for OOM

  #3183 closed Jun 26, 2025

• grpo训练如何设置vllm_device使用多张卡

  #3098 closed Jun 26, 2025

• Does ms-swift support tensor(model)-parallel GRPO training?

  #3068 closed Jun 26, 2025

• ValueError: Image features and image tokens do not match: tokens: 5589, features 5805

  #2460 closed Jun 26, 2025

• grad_norm nan

  #2280 closed Jun 26, 2025

• 期望RLHF能支持序列并行（sequence_parallel）

  #1958 closed Jun 26, 2025

• GRPO训练的old_per_token_logps计算是不是有bug

  #4727 closed Jun 26, 2025

• rerank 数据加载错误

  #4728 closed Jun 26, 2025

• Issue with Multi-GPU Training

  #4718 closed Jun 26, 2025

• Qwen3 Full Sft设置predict_with_generate=true报错keyerror"messages"，为false时可以正常训练结束

  #4695 closed Jun 26, 2025

• 支持 moonshotai/Kimi-VL-A3B-Thinking-2506

  #4708 closed Jun 25, 2025

• grpo训练qwen2.5-vl报错

  #4364 closed Jun 25, 2025

• 全量微调grpo 相同数量的样本ms-swift效果比unsloth效果差很多

  #4393 closed Jun 25, 2025

• GRPO OOM USE resume_from_checkpoint

  #4406 closed Jun 25, 2025

• GRPO训练报错：AssertionError: Forward context is not set. Please use `set_forward_context` to set the forward context.

  #4418 closed Jun 25, 2025

• 支持的DeepSeek-R1训练是指671B的模型吗还是蒸馏的模型？

  #3132 closed Jun 25, 2025

• seq_cls训练时候开启flash_attn指标大幅度低于不开flash_attn

  #4384 closed Jun 25, 2025

• 多回归任务，推理问题

  #4705 closed Jun 25, 2025

• 请问使用zero2/zero3导致max_steps相差八倍的原因是什么？

  #4616 closed Jun 23, 2025

• 请求增加对Qwen3-8B的自我认知训练的NoteBook文件

  #4034 closed Jun 23, 2025

• raise KeyError(f"Column {key} not in the dataset. Current columns in the dataset: {columns}") [rank1]: KeyError: 'Column length not in the dataset. Current columns in the dataset: []'

  #4058 closed Jun 23, 2025

• InternVL3-9B LoRA微调数据集预处理速度缓慢问题（大约7h）

  #4076 closed Jun 23, 2025

• 单坐标点定位物体位置

  #4292 closed Jun 23, 2025

• data_load

  #4288 closed Jun 23, 2025

• Seq CLS Infer 问题咨询

  #4325 closed Jun 23, 2025

• UI-TARS冻结参数推理无法均匀分配显存导致超出显存

  #4359 closed Jun 23, 2025

• 微调Qwen3在默认脚本上加上zero2/3会OOM

  #4371 closed Jun 23, 2025

• VLLM Engine Batch 推理咨询

  #4386 closed Jun 23, 2025

• swift infer这个些命令如何转为python命令运行的，内部原理

  #4555 closed Jun 23, 2025

• Failing to preprocess hf dataset

  #4564 closed Jun 23, 2025

• How-to use on Apple Mac?

  #4572 closed Jun 23, 2025

• Multimodal finetune llava1.6-mistral bug: RuntimeError: Tensors must have same number of dimensions

  #4578 closed Jun 23, 2025

• 关于ms-swift 3.x的template和2.x的不同

  #4602 closed Jun 23, 2025

• ovis2 微调失败，loss计算时报ValueError: Expected input batch_size (1384) to match target batch_size (16384)

  #4611 closed Jun 23, 2025

• 关于pip install -e '.[all]' 的安装、evalscope的安装的咨询

  #4605 closed Jun 23, 2025

• loss_scale hermes not work

  #4607 closed Jun 23, 2025

• 华为910B lora qwen2.5vl报错：AssertionError: Torch not compiled with CUDA enabled

  #4619 closed Jun 23, 2025

• 满血版R1/Qwen3-235B-30A HF参数转megatron OOM

  #4648 closed Jun 23, 2025

• 10分钟改变大模型自我认知教程报错'Qwen2_5VLTemplate' object has no attribute 'model'

  #4662 closed Jun 23, 2025

• DPO训练到 100 步时，遇到 StopIteration ERROR during training 问题

  #4644 closed Jun 23, 2025

• 多卡多进程使用orpo卡死，触发watchdog caught collective operation timeout.

  #3564 closed Jun 20, 2025

• 无

  #4481 closed Jun 20, 2025

• 无

  #4504 closed Jun 20, 2025

• 我希望在训练reward model的时候添加一个分类损失应该怎么做

  #4640 closed Jun 20, 2025

53 Issues opened by 45 people

• 微调DeepSeek模型报错：AssertionError: noaux_tc not supported for training

  #4737 opened Jun 26, 2025

• Does the packing feature block attention score between different samples?

  #4736 opened Jun 26, 2025

• a question for rl

  #4735 opened Jun 26, 2025

• Please open Security Advisories for vulnerability reporting

  #4733 opened Jun 26, 2025

• 在学习全部轮次的SFT训练中，中间轮次结束符号不能被学习，导致训练后的模型无法停止

  #4732 opened Jun 26, 2025

• swift推理精度差异

  #4726 opened Jun 26, 2025

• 使用lora 训练qwen2.5vl3b之后，lora未合并，使用deploy部署，使用pt, 跟vllm 结果不一致

  #4725 opened Jun 26, 2025

• GKD代码加载模型卡死

  #4724 opened Jun 26, 2025

• Swift代码库进行lora checkpoint的continue sft，加载模型和checkpoint后可训练参数为0%

  #4723 opened Jun 26, 2025

• qwen2.5vl lora sft关于freeze_vit

  #4722 opened Jun 26, 2025

• [rank4]: AssertionError: Expected multimodal embeddings to be a list/tuple of 2D tensors, or a single 3D tensor, but got <class 'NoneType'> instead.

  #4721 opened Jun 25, 2025

• qwen3 embedding 微调在评估阶段报错：'NoneType' object has no attribute 'get'

  #4720 opened Jun 25, 2025

• 添加python示例代码

  #4717 opened Jun 25, 2025

• 如何传入自定义的causal_attention_mask

  #4716 opened Jun 25, 2025

• local_repo_path参数，在python脚本里如何添加

  #4714 opened Jun 25, 2025

• hf格式模型文件转megatron报错: CUDA error: operation not supported

  #4713 opened Jun 25, 2025

• lora 微调 Ovis2-34B loss=0.0

  #4711 opened Jun 25, 2025

• Deepspeed zero3 多 GPU 训练没法设置 batch_size 为1

  #4710 opened Jun 25, 2025

• [Bug]: [WARNING:swift] Please install the package: pip install "decord" -U

  #4709 opened Jun 25, 2025

• 多回归任务 输出问题

  #4706 opened Jun 25, 2025

• 序列分类任务，能否多卡训练？

  #4704 opened Jun 25, 2025

• Swift rollout卡住

  #4703 opened Jun 25, 2025

• NuminaMath-TIR数据集上训GRPO不work

  #4702 opened Jun 25, 2025

• DPO的full微调后Qwen3-4B模型不再输出think

  #4701 opened Jun 25, 2025

• 使用msswift框架，基于QwQ-32B模型，微调自制的function-call数据集，效果很差，不知道原因

  #4700 opened Jun 25, 2025

• 自定义数据集包含了'messages'、'rejected_response'、'label'、'images'、'videos'、'audios'、'tools'和'objects'之外的key，该如何写template？

  #4699 opened Jun 24, 2025

• lora微调qwen3 embedding模型弹出警告find_unused_parameters

  #4698 opened Jun 24, 2025

• Qwen2-VL merge lora报错

  #4697 opened Jun 24, 2025

• 求问 Qwen 235B A22 训练成本和 Qwen 32B dense 对比

  #4696 opened Jun 24, 2025

• 用CLI推理时，有办法能在推理结果中保存输入的dataset中的额外参数嘛？

  #4693 opened Jun 24, 2025

• VLLM Engine 咨询

  #4692 opened Jun 24, 2025

• 多机加载大数据集时，会多台机子先后串行加载

  #4691 opened Jun 24, 2025

• UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown

  #4689 opened Jun 24, 2025

• 我想要给PPO设置两个reward model和两个value model，通过两者的value和reward加权计算loss损失，应该怎么做？

  #4688 opened Jun 24, 2025

• 请问是否支持QWenVL等多模态模型的增量预训练？

  #4686 opened Jun 24, 2025

• 改变 IMAGE_FACTOR 是不是意味着视觉部分需要重新训练？

  #4685 opened Jun 24, 2025

• 如何关闭自动模型并行呢？

  #4684 opened Jun 24, 2025

• Help: Multi Turn SFT

  #4681 opened Jun 24, 2025

• mllm模型训练，一个epoch训练完任务卡住，gpu利用率100%，无法save checkpoint

  #4680 opened Jun 23, 2025

• GRPO训练失败，模型似乎学习困难

  #4679 opened Jun 23, 2025

• 奖励函数一直震荡不上升，似乎学不到东西

  #4677 opened Jun 23, 2025

• SFT训练一个回归任务后，推理使用vllm加速，模型load会报错，有办法解决吗

  #4676 opened Jun 23, 2025

• 持续输出Trainer.tokenizer is now deprecated. You should use Trainer.processing_class instead.，但是不报错，请问是什么原因？

  #4673 opened Jun 23, 2025

• 关于 rlhf 数据的preprocess

  #4670 opened Jun 23, 2025

• 微调 MiniCPM-o-2_6 报错 assert media_type in {'image', 'video'}

  #4668 opened Jun 23, 2025

• Applying sequence parallelism causes the training to finish early, even though it hasn't reached the specified max_steps

  #4663 opened Jun 23, 2025

• Any way to run evaluation before training starts?

  #4660 opened Jun 22, 2025

• 'weight' must be 2-D

  #4656 opened Jun 20, 2025

• grpo 多任务训练 奖励函数设置返回None 这样的话，如果想要查看单个任务的reward曲线，在tensorboard中会出现nan的情况

  #4653 opened Jun 20, 2025

• 为什么没有loss

  #4652 opened Jun 20, 2025

• 训练Omni的时候会卡住不动

  #4651 opened Jun 20, 2025

• 请问下swift中集成的lora-ga是否支持多卡训练呢

  #4646 opened Jun 20, 2025

• 更新以后我应该如何获得history呢

  #4645 opened Jun 20, 2025

27 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• qwen2.5vl定位训练同样的参数环境多次训练结果波动非常大

  #4632 commented on Jun 20, 2025 • 0 new comments

• Qwen3 Embedding训练抛出多线程的错误

  #4617 commented on Jun 20, 2025 • 0 new comments

• multi-gpu GRPO training with sequence parallelism terminated unexpectedly during execution, with the following warning: destroy_process_group() was not called before program exit, which may lead to resource leaks.

  #4643 commented on Jun 20, 2025 • 0 new comments

• 怎样为Qwen2.5-VL的视觉和文本设置不同的 lora rank？

  #4223 commented on Jun 20, 2025 • 0 new comments

• 请问会支持qwen2.5中hermes的function call的训练方式吗？

  #3523 commented on Jun 20, 2025 • 0 new comments

• GRPO 训练到100步，评估保存的位置报错，还未进行评估

  #4598 commented on Jun 20, 2025 • 0 new comments

• math_verify 解决表达式能力有限？

  #4541 commented on Jun 20, 2025 • 0 new comments

• pretrain报错进度异常问题

  #2692 commented on Jun 20, 2025 • 0 new comments

• Qwen2.5-vl预训练过程中loss突然激增

  #4634 commented on Jun 23, 2025 • 0 new comments

• Fatal Python error: none_dealloc: deallocating None

  #4353 commented on Jun 23, 2025 • 0 new comments

• lora微调后merge完模型进行lmdeploy推理用时比Qwen2.5-VL-7B-Instruct多一倍，原因为何？

  #4609 commented on Jun 23, 2025 • 0 new comments

• 能否支持MiniCPM-o 2.6 audio模态训练

  #2961 commented on Jun 23, 2025 • 0 new comments

• 采用swift infer 测试qwen2.5-omni模型结果，与官方测试方法结果不一致

  #4595 commented on Jun 23, 2025 • 0 new comments

• qwen3-32B全参数ppo训练一步报错

  #4599 commented on Jun 23, 2025 • 0 new comments

• qwen2.5-7B GRPO训练时卡住，未显示任何报错

  #4603 commented on Jun 23, 2025 • 0 new comments

• 可以在moe的模型训练中 增加专家并行的参数吗

  #1631 commented on Jun 24, 2025 • 0 new comments

• Any example on training llama on function calling dataset?

  #4604 commented on Jun 24, 2025 • 0 new comments

• lora 微调 ovis2-34B loss=0.0 grad_norm=nan

  #3494 commented on Jun 25, 2025 • 0 new comments

• Error occurred when saving checkpoints during Qwen3 multi-GPU SFT

  #4411 commented on Jun 25, 2025 • 0 new comments

• GRPO的时候怎么保存最后一步的checkpoints

  #4574 commented on Jun 25, 2025 • 0 new comments

• 🍭[Roadmap] ms-swift3.6

  #4561 commented on Jun 25, 2025 • 0 new comments

• 训练保存checkpoint的时候报错，但本地又有相应的文件。

  #3420 commented on Jun 25, 2025 • 0 new comments

• 🚀 Best Practices for Training Qwen3/Qwen3-MoE

  #4030 commented on Jun 25, 2025 • 0 new comments

• 支持GME微调么

  #3019 commented on Jun 25, 2025 • 0 new comments

• 想问下embedding的训练如何加入system or instructions？

  #4638 commented on Jun 26, 2025 • 0 new comments

• swift infer 设置了temperature，top_p 但是每次生成都是同样的结果

  #4627 commented on Jun 26, 2025 • 0 new comments

• 训练后的RM模型，支持推理引擎sglang/vllm部署

  #3610 commented on Jun 26, 2025 • 0 new comments