swift issues

InternVL-1.5 希望可以支持

7

https://huggingface.co/OpenGVLab/InternVL-Chat-V1-5 希望可以支持一下微调，实际使用下来效果确实很好

more models

**Describe the feature** Please consider adding support for ORPO training. It extends preference optimization into sft stage and shows promising results. **Paste any useful information** [Paper](https://arxiv.org/abs/2403.07691) and [Support from Huggingface...

zodiacg

多模态 minicpmv2的训练，使用的尺寸是多大，原始minicpm v-2的训练尺寸可以很大，查看源码只看到minicpmv2的输入尺寸是448，训练和推理时对于大图会在原图上进行处理吗，还是会统一直接resize缩放到448？

2

多模态 minicpmv2的训练，使用的尺寸是多大，原始minicpm v-2的训练尺寸可以很大，查看源码只看到minicpmv2的输入尺寸是448，训练和推理时对于大图会在原图上进行处理吗，还是会统一直接resize缩放到448？

zhly0

DDP `include_num_input_tokens_seen` fine-tuning hangs (nccl)

1

Jintao-Huang

good first issue

如何单卡A100推理和微调Qwen1.5-110B-Chat-GPTQ-Int4.

2

Jintao-Huang

good first issue

[WIP]Refactor sequence parallel

# PR type - [ ] Bug Fix - [x] New Feature - [ ] Document Updates - [ ] More Models or Datasets Support # PR information Refactor the...

tastelikefeet

[TorchAcc][Experimental] Integrate more model in torchacc

# PR type - [ ] Bug Fix - [x] New Feature - [ ] Document Updates - [ ] More Model or Dataset Support # PR information Previous PR:...

Zhikaiiii

lisa可训练参数计算异常

2

**Describe the bug** yi-6B-chat可训练参数在`lisa_activated_layers设置1/2/4都显示`[INFO:swift] LlamaForCausalLM: 6061.0355M Params (524.2921M Trainable [8.6502%]), 33.5565M Buffers.`，这个计算是不是有问题？

WSC741606