10 Pull requests merged by 2 people

• [assets] update wechat

  #7019 merged Feb 20, 2025

• Fix Message process of MiniCPMV

  #6998 merged Feb 19, 2025

• [misc] update css

  #6985 merged Feb 18, 2025

• [data] add r1 distill dataset

  #6983 merged Feb 18, 2025

• [version] support transformers 449

  #6982 merged Feb 18, 2025

• [misc] fix script

  #6977 merged Feb 18, 2025

• [data] update vlm args

  #6976 merged Feb 17, 2025

• [data] add min resolution option

  #6975 merged Feb 17, 2025

• [data] fix predict dataset

  #6972 merged Feb 17, 2025

• [assets] update wechat

  #6963 merged Feb 17, 2025

52 Issues closed by 7 people

• 如何在merge lora的时候选择某一个checkpoint

  #7039 closed Feb 23, 2025

• SFT支持text形式的数据格式吗?

  #7029 closed Feb 21, 2025

• 是否支持Megatron-Deepspeed的模型并行？

  #7033 closed Feb 21, 2025

• rm训练

  #7031 closed Feb 21, 2025

• format_prefix 使用问题

  #7026 closed Feb 21, 2025

• webui多卡VLLM推理错误

  #6468 closed Feb 21, 2025

• ChildFailedError for qwenvl2

  #7032 closed Feb 21, 2025

• SFT

  #7028 closed Feb 21, 2025

• qwen 的数据处理模板只剩下 一个 qwen 了吗？

  #7025 closed Feb 21, 2025

• 为什么打开webui界面加载不出来，模型使用的是deepseekr1 1.5BAWQ版本。

  #7023 closed Feb 21, 2025

• No template fot qwen_2_5_vl

  #7024 closed Feb 21, 2025

• ValueError: Some specified arguments are not used by the HfArgumentParser: ['vicuna']

  #7021 closed Feb 21, 2025

• 为什么我微调的时候，显存使用情况与官网提供的列表完全不同？

  #7022 closed Feb 21, 2025

• minicpm_o Batch inference test set，

  #7011 closed Feb 21, 2025

• 模型相关

  #7018 closed Feb 20, 2025

• 参数相关

  #7017 closed Feb 20, 2025

• rm模型代码更改问题

  #6967 closed Feb 20, 2025

• 请问 qwen2.5vl 做图像目标检测 vision grounding 任务，如何构建finetune 数据集

  #7015 closed Feb 20, 2025

• RM使用AutoModelForSequenceClassification 加载问题

  #7012 closed Feb 20, 2025

• 单机多卡微调qwen.25-VL模型，微调加载模型后卡住不动，单卡微调训练没有问题。

  #6990 closed Feb 20, 2025

• 关于special token的问题

  #7007 closed Feb 20, 2025

• 关于ppo训练问题

  #7008 closed Feb 20, 2025

• Evaluate and Predict Error in webui

  #7009 closed Feb 20, 2025

• kto_en_demo error

  #7005 closed Feb 19, 2025

• How to use llama factory with unsloth for multi GPU support?

  #6940 closed Feb 19, 2025

• 导出模型时，加载的.safetensors后缀的文件是属于哪一个checkpoint文件夹

  #7003 closed Feb 19, 2025

• Webui chat 总是乱码

  #7002 closed Feb 19, 2025

• 微调minicpm-o-2.6报错（语音输入，文本输出多轮微调。

  #6973 closed Feb 19, 2025

• 请问现在支持vl模型lora微调后直接使用adapter进行推理吗？

  #6993 closed Feb 19, 2025

• ppo和dpo数据集格式问题

  #6994 closed Feb 19, 2025

• 有哪些参数会使得数据预处理后examples数量显著减少？

  #6991 closed Feb 19, 2025

• 自己微调出蒸馏模型，有增加<think>...</think>的标志吗

  #6996 closed Feb 19, 2025

• 请问对于这种格式Congliu/Chinese-DeepSeek-R1-Distill-data-110k的训练数据，是不是使用的template值是deepseek3

  #6997 closed Feb 19, 2025

• 大佬，啥时候支持ds的GPRO

  #6995 closed Feb 19, 2025

• DeepSeek-R1-Distill-Qwen- sft时候chat_template不一致

  #6987 closed Feb 18, 2025

• 导出模型时，加载的.safetensors后缀的文件是属于哪一个checkpoint文件夹

  #6979 closed Feb 18, 2025

• deepseek V3满血版推理

  #6978 closed Feb 18, 2025

• 安装完依赖包后，执行llamafactory-cli help 命令报错

  #6974 closed Feb 17, 2025

• 基于qwen25vl_7b_instruct lora微调后的模型推理报错KeyError: 0

  #6960 closed Feb 17, 2025

• 计算 BLEU 和 ROUGE 分数的predict报错

  #6952 closed Feb 17, 2025

• 运行llama3-8b推理，模型总是胡言乱语，并且一直重复输出

  #6969 closed Feb 17, 2025

• 保存checkpoint之后的下一个step loss为0

  #6966 closed Feb 17, 2025

• 微信群加上不了 提示需要群成员邀请

  #6956 closed Feb 17, 2025

• Heartbeat monitor timed out! Process will be terminated after dumping debug info. workMetaList_.size()=1

  #6953 closed Feb 17, 2025

• 模型输出停不下来，最后没有输出停止符。

  #6951 closed Feb 17, 2025

• GPU使用率高、显存使用率低、内存使用高

  #6962 closed Feb 17, 2025

• lora sft qwen2.5-vl-7B transformers version bug

  #6968 closed Feb 17, 2025

• Qwen2.5-VL *B模型 SFT训练&使用vllm推理报错：AttributeError: 'NoneType' object has no attribute 'image_processor'

  #6965 closed Feb 17, 2025

• Qwen2.5-VL 在相同的conda环境下，webui上的推理结果和用transformers推理的结果差异较大

  #6925 closed Feb 17, 2025

• transformers for qwen-2_5_vl has updated

  #6941 closed Feb 17, 2025

• 长时间输出会变成英文编程？

  #6961 closed Feb 17, 2025

• Qwen2-VL 多图 SFT

  #6959 closed Feb 17, 2025

19 Issues opened by 19 people

• Inability to effectively fine-tune models with built-in inference capabilities

  #7042 opened Feb 23, 2025

• Long context full SFT validation causes OOM

  #7041 opened Feb 23, 2025

• colab和kaggle平台部署报错

  #7037 opened Feb 22, 2025

• Unable to use plain text for chat or eval of multimodal model (ValueError: Invalid input type. Must be a single image, a list of images, or a list of batches of images.)

  #7036 opened Feb 22, 2025

• 基于最新的LLaMA-Factory训练Qwen2.5-vl, 训练变得非常慢

  #7030 opened Feb 21, 2025

• deepseek r1微调

  #7027 opened Feb 21, 2025

• UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc0 in position 17: invalid start byte

  #7016 opened Feb 20, 2025

• 使用昇腾910A llama factory sft qwen2-7b时报错E40024: 2025-02-20-14:05:46.947.014 Failed call Python Func/Meathod [get_binfile_sha256_hash_from_c],

  #7014 opened Feb 20, 2025

• 在服务器端配置端口映射后，WebUI无法正常显示。确实前端css文件

  #7013 opened Feb 20, 2025

• 使用API模型能力变差

  #7010 opened Feb 20, 2025

• mac本用llama-factory如何用MPS ？

  #7001 opened Feb 19, 2025

• A800 7*80g 全参微调qwen-2.5-32b OOM？

  #6999 opened Feb 19, 2025

• 启动WebUI失败

  #6989 opened Feb 18, 2025

• minicpm_2_6o全量微调验证集eval_loss不计算不打印，也不绘制eval_loss图

  #6988 opened Feb 18, 2025

• [help]如何添加规则到数据集的每个条目中，并且不影响返回值

  #6984 opened Feb 18, 2025

• 怎么返回输出token的prob呢?

  #6980 opened Feb 18, 2025

• converting model Error: unknown data type: I32

  #6971 opened Feb 17, 2025

• 用deepspeed zero-3-offload去微调DeepSeek-R1-Distill-Qwen-32B，系统卡住，长时间无反应

  #6964 opened Feb 17, 2025

• [HELP] Runable solution of RTX 5090 GPU + Linux Driver version + Pytorch version + Deepspeed version for LLM finetuning

  #6958 opened Feb 17, 2025

21 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.

• One of the subprocesses has abruptly died during map operation.To debug the error, disable multiprocessing. 无论numworkers多大都没用

  #6762 commented on Feb 17, 2025 • 0 new comments

• NPU ds3_ofld训练不释放内存最终OOM

  #6816 commented on Feb 17, 2025 • 0 new comments

• Serious misalignment in LLaVA implementation

  #6008 commented on Feb 18, 2025 • 0 new comments

• Qwen2.5-VL full sft dtype error

  #6791 commented on Feb 18, 2025 • 0 new comments

• 有计划支持Deepseek的janus pro微调么

  #6775 commented on Feb 18, 2025 • 0 new comments

• Support for finetuning the convolutional layers of VLMs with lora

  #6869 commented on Feb 18, 2025 • 0 new comments

• 训练PPO出现问题：ValueError: Target module ModuleDict( (default): Identity() (reward): Identity() ) is not supported. Currently, only the following modules are supported: `torch.nn.Linear`, `torch.nn.Embedding`, `torch.nn.Conv2d`, `transformers.pytorch_utils.Conv1D`.

  #6373 commented on Feb 19, 2025 • 0 new comments

• ppo训练出现问题

  #6881 commented on Feb 19, 2025 • 0 new comments

• Error in the process of fine-tuning qwen2.5vl-7b evaluate&predict data = [self.dataset[idx] for idx in possibly_batched_index] KeyError: 0

  #6947 commented on Feb 19, 2025 • 0 new comments

• llamafactory最新版0.9.2.dev0，unsloth加速训练报错

  #6836 commented on Feb 19, 2025 • 0 new comments

• 结合 GRPO 支持 DeepSeek-R1 等推理模型的复现，达到 huggingface open-r1 的类似效果

  #6792 commented on Feb 20, 2025 • 0 new comments

• Qwen2.5-VL-7B-Instruct推理bug：RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1!

  #6910 commented on Feb 20, 2025 • 0 new comments

• llamafactory-cli api 使用vllm与直接使用vllm serve模型效果不一致

  #5409 commented on Feb 20, 2025 • 0 new comments

• 能否支持在Qlora微调后直接导出int4等量化的模型

  #6747 commented on Feb 21, 2025 • 0 new comments

• 求助：如何对 DeepSeek R1 进行 SFT

  #6824 commented on Feb 21, 2025 • 0 new comments

• 单机多卡 resume_from_checkpoint 时报错 assert len(self.ckpt_list) > 0

  #6955 commented on Feb 21, 2025 • 0 new comments

• 请问是否会在框架内集成RLOO算法，最新的online RLHF？

  #4287 commented on Feb 21, 2025 • 0 new comments

• 最新版本 trl PPOConfig 不兼容

  #5936 commented on Feb 21, 2025 • 0 new comments

• failed to docker build

  #6922 commented on Feb 23, 2025 • 0 new comments

• add Sequence Parallelism

  #6506 commented on Feb 22, 2025 • 0 new comments

• DeepSeekV3-671B-BF16 Lora Finetune

  #6843 commented on Feb 23, 2025 • 0 new comments