Repository Issue Activity (beta)

#10757 ppo报错

#10758 upstream-sync: rebase conflict 2026-08-17

#10709 top_p=0.0 is silently rewritten to 1.0 before reaching vLLM, inverting the request

#8982 Training using CPU only?

#10756 Prevent Margin Leaks & Surprise LLM Bills: Add a Hybrid Spend Firewall

#10755 [Bug] DPO training crashes with Qwen3.5-35B-A3B (MoE VLM): TypeError: expected Tensor as element 0 in argument 0, but got list in collator

#10740 about ppo unwrap_model

#10646 Unauthenticated SSRF via image_url in the API server, bypassing check_ssrf_url by redirect and DNS rebinding (incomplete fix)

#10736 packing 在 SDPA 路径下破坏 rotary embedding(Gemma-3/4, head_dim=256):position_ids 未按 pad_to_multiple_of 补齐

#10670 [Bug] Packing mode: right_padding_length may exceed valid range when max_cutoff_len is improper

#10732 Python 3.14 support addition

#10721 RFC: Expert Parallel Training in LlamaFactory V1 with FSDPTurbo

#10679 Can anybody explain what is v1?

#10704 单机多卡，开启swanlab日志输出，报错：[rank0]: raise RuntimeError("No active Run. Call swanlab.init() first.")

#10647 Gemma4-12B Training Failed

#10474 RFC: Ulysses Context Parallelism for LlamaFactory v0

#6429 寒武纪：咱们是否能支持寒武纪？

#10682 Add REFUTE scientific critique + calibration benchmark

#9567 Qwen2.5-1.5B 进行SFT训练 grad_norm固定为1.73205 loss一开始很大后面降为0

#6373 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`.

#5484 请问DPO训练的时候有什么注意事项吗？我训练出来效果很差。

#10675 _calculate_timestamps参数使用

#10673 Feature Request: Support semantic-aware sequence packing for improved training efficiency

#10672 Feature Request: Support per-epoch resampling for interleave dataset mixing strategy

#10669 [Bug] Qwen2.5-Omni: Missing video image lengths in RoPE position computation