Discussions

• 1 You must be logged in to vote

  🙏

  Qwen3-30b-a3b lora训练结束后保存最终checkpoint失败

  alex-wll asked Sep 15, 2025 in Q&A · Unanswered

  

  

  1
• 1 You must be logged in to vote

  💬

  llama factory 的chat 推理结果和vllm 推理结果个别地方不一样

  johnsonwag03 started Jan 26, 2026 in General

  

  

  0
• 1 You must be logged in to vote

  🙏

  qwen3-30b-thinking在32G*8的华为昇腾上微调OOM？

  gongyinfeng0206-design asked Jan 23, 2026 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  🙏

  How to fine tune Qwen3 VL model for grounding task

  WiedenWei asked Jan 21, 2026 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  🙏

  Qwen3模型训练，训练数据混合了CoT和非CoT数据，应该怎么设置

  GaryZhu1996 asked Aug 6, 2025 in Q&A · Unanswered

  

  

  

  

  3
• 1 You must be logged in to vote

  🙏

  全参数训练怎么导出为safetensor格式

  AragornHorse asked Jan 17, 2026 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  💬

  如果采用的不是constant学习率方法，那么最低学习率应该如何设置？目前好像只能设置最高学习率，最低学习率为0在大多场景好像效果并不好

  bug Something isn't working pending This problem is yet to be addressed

  DreamDimo started Jan 15, 2026 in General

  

  

  0
• 1 You must be logged in to vote

  🙏

  多轮对话的微调Funnction calling 如何评估？仓库中有代码？

  pending This problem is yet to be addressed

  cqray1990 asked Jan 15, 2026 in Q&A · Unanswered

  

  

  0
• 2 You must be logged in to vote

  🙏

  关于使用LLama-factory进行tool_call训练的正确姿势

  Fueav asked Sep 5, 2025 in Q&A · Unanswered

  

  

  1
• 4 You must be logged in to vote

  🙏

  ppo和dpo数据集格式问题

  wontfix This will not be worked on

  Youmo1 asked Feb 19, 2025 in Q&A · Unanswered

  

  

  1
• 1 You must be logged in to vote

  🙏

  DPO训练多轮对话数据时格式设置问题

  Ch4tppt asked Jul 3, 2025 in Q&A · Unanswered

  

  

  1
• 1 You must be logged in to vote

  🙏

  请问DPO的多轮对话格式数据集中可以支持<image>多模态吗？

  Yukinonooo asked Jan 14, 2026 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  🙏

  qwen3-vl使用DPO训练过程loss降到0了，但是训练后的模型推理很奇怪

  bug Something isn't working pending This problem is yet to be addressed

  Xinyu-lab asked Jan 12, 2026 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  🙏

  Image input with Gemma3 fine-tuning

  chrismrutherford asked Jan 10, 2026 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  🙏

  请问现在除了 nvidia 和华为昇腾的卡，还有对其他国产厂商的训练支持吗，比如天数智芯，壁仞，沐曦等

  SeaSoonKeun asked Jan 9, 2026 in Q&A · Unanswered

  

  

  0
• 2 You must be logged in to vote

  🙌

  llamafactory中用lora微调qwen3-32b-awq模型时梯度为nan损失为0的问题及其解决方案

  gysabc started Sep 12, 2025 in Show and tell

  

  

  3
• 1 You must be logged in to vote

  🙏

  Qwen3-VL 8B Training Running excessively Slow

  Kathan-Vakharia asked Dec 22, 2025 in Q&A · Answered

  

  

  1
• 2 You must be logged in to vote

  🙏

  Version conflict issue with TRL during PPO training.

  HuDi2018 asked Jan 4, 2026 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  🙏

  有训练Qwen3-Omni的.yaml配置示例嘛？

  gauss-clb asked Dec 30, 2025 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  🙏

  两张4090，加载数据集反而比一张4090慢了5-6倍

  Sleep-Enough asked Dec 29, 2025 in Q&A · Unanswered

  

  

  0
• 5 You must be logged in to vote

  🙏

  Qwen2.5-VL全参数训练视觉层和projector，lora微调语言模型的模型训练与导出方式？

  bug Something isn't working pending This problem is yet to be addressed

  echochoc asked Mar 18, 2025 in Q&A · Unanswered

  

  

  1
• 1 You must be logged in to vote

  🙏

  packing 时卡间attention计算速度存在显著差异

  JL-Cheng asked Dec 11, 2025 in Q&A · Unanswered

  

  

  0
• 1 You must be logged in to vote

  🙏

  是在是很想知道 为什么DPO训练的效果如此异常， 谢谢各位大佬，请您点拨一下

  652994331 asked Jan 21, 2025 in Q&A · Unanswered

  

  

  

  

  5
• 1 You must be logged in to vote

  💬

  微调 qwen3-0.6B template

  xiaojun3619 started Dec 10, 2025 in General

  

  

  0
• 1 You must be logged in to vote

  🙏

  llama factory支持openAI格式的多模态微调么

  hwang136 asked Dec 10, 2025 in Q&A · Unanswered

  

  

  0