4 Releases published by 1 person

• v2.3.2

  published Aug 24, 2024

• v2.4.0

  published Sep 13, 2024

• v2.4.1

  published Sep 13, 2024

• v2.4.2

  published Sep 18, 2024

129 Pull requests merged by 11 people

• fix qwen2.5 template

  #2081 merged Sep 20, 2024

• dynamic vit gradient_checkpointing

  #2071 merged Sep 20, 2024

• Support Mistral-small-inst-2409

  #2077 merged Sep 20, 2024

• fix RLHF & max_length

  #2075 merged Sep 19, 2024

• Update qwen2-vl最佳实践.md

  #2058 merged Sep 19, 2024

• fix rlhf zero3

  #2072 merged Sep 19, 2024

• Fix yi template

  #2067 merged Sep 19, 2024

• fix win32 quote

  #2065 merged Sep 18, 2024

• update qwen2-vl docs

  #2063 merged Sep 18, 2024

• fix notebook gradio

  #2062 merged Sep 18, 2024

• support qwen2.5-coder

  #2061 merged Sep 18, 2024

• vllm support mutli image

  #2059 merged Sep 18, 2024

• support qwen2-vl -72b/qwen2.5-math/qwen2.5-coder

  #2056 merged Sep 18, 2024

• Support qwen2.5

  #2054 merged Sep 18, 2024

• support qwen2-vl-base

  #2052 merged Sep 18, 2024

• fix qwen2vl position_ids

  #2051 merged Sep 18, 2024

• update docs

  #2050 merged Sep 17, 2024

• llama3 tool calling

  #2048 merged Sep 15, 2024

• Fix multi coordinate grounding

  #2047 merged Sep 15, 2024

• support multi bbox grounding

  #2045 merged Sep 15, 2024

• fix mplug-owl3

  #2042 merged Sep 14, 2024

• Add longwriter filtered dataset

  #2037 merged Sep 14, 2024

• fix rlhf & zero3

  #2034 merged Sep 14, 2024

• Fix olora and pissa saving files which will cause the second saving failed

  #2032 merged Sep 13, 2024

• fix deploy eval kill

  #2029 merged Sep 12, 2024

• update code

  #2028 merged Sep 12, 2024

• refactor rlhf

  #1975 merged Sep 12, 2024

• Florence use _post_encode & template support encoder-decoder

  #2019 merged Sep 11, 2024

• Add FAQ Document

  #2013 merged Sep 11, 2024

• fix lmdeploy qwen_vl

  #2009 merged Sep 11, 2024

• Support llava1.6-llama3.1-8b-instruct

  #2005 merged Sep 10, 2024

• Fix rlhf ref model

  #2003 merged Sep 10, 2024

• compat lmdeploy==0.6

  #2001 merged Sep 10, 2024

• fix EngineGenerationConfig importError of lmdeploy

  #1990 merged Sep 10, 2024

• Support Deepseek 2.5

  #1992 merged Sep 10, 2024

• fix

  #1995 merged Sep 10, 2024

• fix patch

  #1997 merged Sep 10, 2024

• fix model_mapping

  #1982 merged Sep 9, 2024

• fix typo

  #1980 merged Sep 9, 2024

• Add reflection model

  #1973 merged Sep 7, 2024

• update docs

  #1970 merged Sep 7, 2024

• support mplug_owl3

  #1957 merged Sep 7, 2024

• fix bugs

  #1959 merged Sep 6, 2024

• Fix the lora hook

  #1963 merged Sep 6, 2024

• Fix data info print in rlhf

  #1964 merged Sep 6, 2024

• Add lazy_tokenize to RLHF

  #1956 merged Sep 6, 2024

• Support minicpm 3

  #1952 merged Sep 5, 2024

• fix rlhf

  #1949 merged Sep 5, 2024

• support dynamic_eos

  #1947 merged Sep 5, 2024

• fix qwen2-vl & video

  #1950 merged Sep 5, 2024

• fix file rename error in megatron when there are multi process

  #1948 merged Sep 5, 2024

• refactor rlhf

  #1885 merged Sep 5, 2024

• fix do_sample

  #1946 merged Sep 5, 2024

• fix lmdeploy seed

  #1945 merged Sep 5, 2024

• update yi-coder

  #1942 merged Sep 5, 2024

• fix swift deploy

  #1936 merged Sep 4, 2024

• fix typing

  #1933 merged Sep 4, 2024

• Support deploy & logprobs

  #1833 merged Sep 4, 2024

• [TorchAcc] fix: fix the judegement of fsdp_num

  #1903 merged Sep 4, 2024

• update docs & fix bug

  #1926 merged Sep 4, 2024

• update wechat

  #1925 merged Sep 3, 2024

• [TorchAcc] perf: use xm.save instead of torch.save

  #1916 merged Sep 3, 2024

• refactor docs

  #1915 merged Sep 3, 2024

• Refactor docs

  #1912 merged Sep 3, 2024

• fix web-ui push to hub strategy

  #1909 merged Sep 3, 2024

• update docs

  #1908 merged Sep 3, 2024

• deepspeed use cosine lr_schduler

  #1907 merged Sep 2, 2024

• support logprobs

  #1900 merged Sep 2, 2024

• fix push_to_ms

  #1901 merged Sep 2, 2024

• support custom quantized dataset

  #1893 merged Sep 2, 2024

• Fix push_to_hub when last-checkpoint

  #1897 merged Sep 2, 2024

• Add some warnings and fix RLHF

  #1890 merged Sep 2, 2024

• add vllm lmdeploy benchmark

  #1889 merged Sep 2, 2024

• Fix push to hub logic

  #1888 merged Sep 2, 2024

• Refactor push_to_hub

  #1883 merged Sep 2, 2024

• support qwen2-vl gptq awq

  #1884 merged Sep 2, 2024

• Support freeze vit

  #1880 merged Aug 31, 2024

• use model.generation_config

  #1850 merged Aug 31, 2024

• add duet

  #1877 merged Aug 31, 2024

• Fix neftune doc

  #1875 merged Aug 31, 2024

• Fix num_proc

  #1874 merged Aug 30, 2024

• Add train record

  #1873 merged Aug 30, 2024

• [TorchAcc] fix serveral bugs for torchacc FSDP.

  #1872 merged Aug 30, 2024

• Support faster data map

  #1871 merged Aug 30, 2024

• update docs qwen2-vl

  #1869 merged Aug 30, 2024

• fix requirements

  #1864 merged Aug 30, 2024

• fix qwen2-vl docs

  #1861 merged Aug 30, 2024

• update qwen2-vl docs

  #1858 merged Aug 29, 2024

• update qwen2-vl docs

  #1856 merged Aug 29, 2024

• Update new datasets

  #1855 merged Aug 29, 2024

• support qwen2-vl & video finetune

  #1849 merged Aug 29, 2024

• Support qwen2 vl grounding

  #1854 merged Aug 29, 2024

• Fix Pissa and OLoRA

  #1852 merged Aug 29, 2024

• Fix some datasets for streaming

  #1848 merged Aug 29, 2024

• Add internvl2 awq models

  #1846 merged Aug 29, 2024

• support qwen2-vl

  #1842 merged Aug 29, 2024

• Support eval_nproc

  #1843 merged Aug 29, 2024

• fix internlm-xcomposer rlhf

  #1838 merged Aug 28, 2024

• add ddp_timeout parameter

  #1836 merged Aug 27, 2024

• support qwen2-pro dataset

  #1834 merged Aug 27, 2024

• fix inject

  #1835 merged Aug 27, 2024

• Fix code

  #1824 merged Aug 27, 2024

• fix minicpm-v 2.6 infer device_map

  #1832 merged Aug 27, 2024

• use default-lora

  #1823 merged Aug 27, 2024

• Support register loss func

  #1822 merged Aug 26, 2024

• fix dora deployment

  #1821 merged Aug 26, 2024

• Support liger

  #1819 merged Aug 26, 2024

• fix preprocess_num_proc

  #1818 merged Aug 26, 2024

• fix mp+ddp & resume_from_checkpoint

  #1815 merged Aug 26, 2024

• Support zero2 offload

  #1814 merged Aug 26, 2024

• compat with vllm==0.5.5

  #1812 merged Aug 25, 2024

• fix

  #1811 merged Aug 23, 2024

• fix offline export

  #1805 merged Aug 23, 2024

• Support Latex OCR dataset

  #1810 merged Aug 23, 2024

• Support hd num

  #1801 merged Aug 23, 2024

• fix megatron_patch_path

  #1804 merged Aug 23, 2024

• fix CI

  #1797 merged Aug 23, 2024

• fix mllm rlhf with full sft type

  #1800 merged Aug 22, 2024

• fix history_roles

  #1798 merged Aug 22, 2024

• fix imports

  #1796 merged Aug 22, 2024

• fix bugs

  #1794 merged Aug 22, 2024

• fix yi-vl template

  #1793 merged Aug 22, 2024

• support qwen-vl & base64

  #1790 merged Aug 22, 2024

• update doc

  #1789 merged Aug 22, 2024

• ReFT

  #1785 merged Aug 21, 2024

• support phi3.5-vision

  #1780 merged Aug 21, 2024

• fix moe & gradient_checkpointing

  #1782 merged Aug 21, 2024

• fix infer dataset_test_ratio

  #1779 merged Aug 21, 2024

• Fix zero3 & minicpm-v/internvl2/xcomposer

  #1772 merged Aug 20, 2024

4 Pull requests opened by 4 people

• Add Internvl2TemplateWithAngles template

  #1784 opened Aug 21, 2024

• # 观察数据后，发现下面的代码会过滤掉一些没有问题的数据，如：sure, here are some tools and …

  #1931 opened Sep 4, 2024

• [WIP]Feat/refactor3

  #2030 opened Sep 12, 2024

• update qwen2.5 best practices

  #2080 opened Sep 20, 2024

283 Issues closed by 30 people

• OOM when tokenizing datasets

  #1971 closed Sep 20, 2024

• DPO training error `RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0!`

  #2044 closed Sep 20, 2024

• Swift DPO Template 格式问题

  #2023 closed Sep 20, 2024

• 模型推理保存路径 (result_path)

  #2026 closed Sep 20, 2024

• qwen2 vl 72B 微调报错

  #2068 closed Sep 20, 2024

• internvl2-8b lora微调后 merge lora出错（if llm_config['architectures'][0] == 'LlamaForCausalLM':）

  #2073 closed Sep 20, 2024

• 请问DPO微调 Prompt部分会计算Loss么？

  #2070 closed Sep 19, 2024

• swift2.4.2版本更新支持qwen2.5之后，但是yi-1.5-34b推理报错(更新之前是支持的)

  #2066 closed Sep 19, 2024

• llava video SFT BUG

  #2060 closed Sep 19, 2024

• windows使用webui训练报错 invalid float value: "'1e-4'"

  #2036 closed Sep 18, 2024

• chatglm4v-9b全量微调full报错：File "/usr/local/Python3.10.0/lib/python3.10/site-packages/deepspeed/runtime/zero/stage3.py", line 149, in __init__ [rank0]: self.dtype = self.optimizer.param_groups[0]['params'][0].dtype [rank0]: IndexError: list index out of range

  #2043 closed Sep 18, 2024

• 换了多个模型，都报一个错误：sh:1:syntax

  #1994 closed Sep 17, 2024

• dpo微调与zero3不兼容

  #1899 closed Sep 17, 2024

• DPO微调InternVL2-2B时报错

  #1979 closed Sep 17, 2024

• eval_acc 是如何计算的？

  #2006 closed Sep 17, 2024

• Does DPO/RLHF tuning support internVL2 video models?

  #2015 closed Sep 17, 2024

• 使用cogvlm2在rlaif-v数据集上做DPO训练报错

  #2025 closed Sep 17, 2024

• Merge Lora后为什么sft_type被设置为full，不应该是lora么

  #2041 closed Sep 17, 2024

• internlm-xcomposer2-7b-chat 使用 use_flash_attn 出现错误

  #2046 closed Sep 17, 2024

• [Re-appeared] DPO training error UnboundLocalError: local variable 'num_patches' referenced before assignment

  #2012 closed Sep 14, 2024

• transformers>=4.45.0.dev0

  #2018 closed Sep 14, 2024

• DPO support resume_from_checkpoint

  #2031 closed Sep 14, 2024

• Merge Lora后为什么stf被设置为full

  #2040 closed Sep 14, 2024

• cannot import name 'ftp_head' from 'datasets.utils.file_utils'

  #2021 closed Sep 12, 2024

• dpo internvl2存在mismatch

  #1930 closed Sep 12, 2024

• Internlm-Xcomposer2.5 推理时输入多张图报错

  #2020 closed Sep 11, 2024

• Uploading models to Hugging Face from MS Swift

  #1981 closed Sep 11, 2024

• Issue: RuntimeError with Multiple GPUs in MS Swift 2.5.0-dev

  #1984 closed Sep 11, 2024

• DPO training error `UnboundLocalError: local variable 'num_patches' referenced before assignment`

  #1734 closed Sep 11, 2024

• O-lora 不可用

  #1989 closed Sep 11, 2024

• dpo训练时，使用自带数据集，报错

  #2008 closed Sep 11, 2024

• lr_scheduler_type

  #1983 closed Sep 10, 2024

• lmdeploy的main分支已经移除了EngineGenerationConfig，目前使用swift调用main分支的lmdeploy会报错

  #1985 closed Sep 10, 2024

• 评测时出现超时错误

  #1974 closed Sep 10, 2024

• 使用deepspeed时，出现RuntimeError: 'weight' must be 2-D 的错误

  #1996 closed Sep 10, 2024

• Can't do RLHF with InternVL2-4B

  #1987 closed Sep 9, 2024

• qwen2-vl-7b-instruct 推理视频报错

  #1988 closed Sep 9, 2024

• BUG：init_lora lead to the wrong distribute？

  #1944 closed Sep 9, 2024

• InternVL2 全量微调时显存占用持续上涨

  #1955 closed Sep 9, 2024

• Huggingface link is broken

  #1976 closed Sep 8, 2024

• Cannot get model_type from the deploy service

  #1904 closed Sep 7, 2024

• 如何手动下载评测数据集，并且如何指定评测时的数据集路径

  #1972 closed Sep 7, 2024

• ImportError: cannot import name 'LlavaOnevisionForConditionalGeneration' from 'transformers'

  #1878 closed Sep 7, 2024

• 环境镜像

  #1966 closed Sep 7, 2024

• support minicpm3-4b

  #1962 closed Sep 6, 2024

• TypeError: BaseAWQForCausalLM.quantize() got an unexpected keyword argument 'n_parallel_calib_samples'

  #1960 closed Sep 6, 2024

• GLM-4V-9B fine-tuning error: ValueError: 151339 is not in list

  #1840 closed Sep 5, 2024

• Qwen2-VL-7B-instruct 微调报错：RuntimeError: CUDA error: too many resources requested for launch

  #1927 closed Sep 5, 2024

• TypeError: Qwen2ForCausalLM.forward() got an unexpected keyword argument '_data'

  #1929 closed Sep 5, 2024

• Qwen2-VL-7B-Instruct Video inference

  #1920 closed Sep 5, 2024

• swift infer的时候，传递do_sample参数不起作用

  #1943 closed Sep 5, 2024

• 两张V100微调qwen2-7b，单卡微调正常，双卡微调出现RuntimeError: CUDA error: device-side assert triggered CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.

  #1882 closed Sep 5, 2024

• 使用自定数据集DPO mllm时报错KeyError: 'prompt'

  #1922 closed Sep 5, 2024

• qwen2-vl-2b-instruct微调报错ValueError: push_best is not a valid HubStrategy, please select one of ['end', 'every_save', 'checkpoint', 'all_checkpoints']

  #1917 closed Sep 5, 2024

• GLM4V-9B微调后，直接调用微调的模型出错：ValueError: The following `model_kwargs` are not used by the model: ['images'] (note: typos in the generate arguments will also show up in this list)

  #1902 closed Sep 5, 2024

• Merge LoRA & 量化部分 支持bnb量化嘛

  #1895 closed Sep 5, 2024

• dpo微调internvl2

  #1886 closed Sep 5, 2024

• 请教一下，图片、视频等多模态数据超长应该怎么截断？

  #1876 closed Sep 5, 2024

• 部分数据集处理时出现了超过max length的warining，但实际数据貌似并没有超过，前几个版本的分支没有这个问题

  #1865 closed Sep 5, 2024

• Qwen2-VL-7B的微调out of memory

  #1860 closed Sep 5, 2024

• Can't find forward pass entrypoint for InternVL2

  #1851 closed Sep 5, 2024

• agent tools工具在调用时思维链已经决定了多个action在tools_call时只返回一个

  #1826 closed Sep 5, 2024

• The setting of the "--device_max_memory" parameter seems to have no effect on the even distribution of GPU memory.

  #1830 closed Sep 5, 2024

• 关于推理时获得logits结果

  #1528 closed Sep 4, 2024

• swift backend使用vllm推理时，openai接口不支持logprobs输出概率

  #1719 closed Sep 4, 2024

• LLaVA-OneVision 的支持

  #1662 closed Sep 4, 2024

• lr_scheduler_type

  #1894 closed Sep 4, 2024

• qwen2-vl-2b-instruct使用自定义数据集微调出现DatasetGenerationError

  #1918 closed Sep 4, 2024

• internvl2-26b多卡训练报错“Expected all tensors to be on the same device...”

  #1910 closed Sep 3, 2024

• qwen2-vl-2b-instruct微调报错：importlib.metadata.PackageNotFoundError: No package metadata was found for The 'qwen_vl_utils' distribution was not found and is required by this application.

  #1905 closed Sep 3, 2024

• AssertionError: DeepSpeed does not recognize LR scheduler WarmupCosineLR

  #1906 closed Sep 2, 2024

• Failed to import swift.llm.sft because of the following error

  #1898 closed Sep 2, 2024

• internvl-40b模型微调后推理时报错

  #1881 closed Aug 31, 2024

• How to freeze the ViT part during full parameter fine-tuning of Qwen2-vl?

  #1879 closed Aug 31, 2024

• zero-3微调internvl2-40B

  #1845 closed Aug 31, 2024

• qwen2-vl fine-tuning error: module 'torch.nn' has no attribute 'RMSNorm'

  #1870 closed Aug 30, 2024

• deepspeed-zero3: Input should be a valid integer, got a number with a fractional part [type=int_from_float, input_value=11560550.4, input_type=float]

  #1862 closed Aug 30, 2024

• Qwen2-VL-7B的微调out of memory

  #1859 closed Aug 30, 2024

• 多模态模型DPO微调时的报错

  #1853 closed Aug 29, 2024

• ValueError: model_type: 'internvl2-8b-awq' is not registered.

  #1847 closed Aug 29, 2024

• Examples of running Inference on AWQ models with swift? I'm trying to run the AWQ version of InternVL2 models

  #1841 closed Aug 29, 2024

• 是否支持Lora继续训练？有报错

  #1828 closed Aug 28, 2024

• 训练数据咨询

  #1827 closed Aug 28, 2024

• 多轮对话conversations只有一组怎么处理？

  #1816 closed Aug 28, 2024

• MiniCPM-V2.6在自己数据上LoRA微调，训练过程中报错DataLoader worker (pid 184751) is killed by signal: Segmentation fault.

  #1806 closed Aug 28, 2024

• 添加packing之后自定义数据集报错KeyError: 'input_ids'，不使用packing可以正常训练

  #1795 closed Aug 28, 2024

• 请问glm4v什么时候支持lmdeploy部署呢？

  #1791 closed Aug 28, 2024

• 感觉是用了fp16的原因. 你可以试试bf16或者fp32不

  #1788 closed Aug 28, 2024

• 为了支持zsh改动后，不支持windows安装

  #1776 closed Aug 28, 2024

• internvl-chat-v1_5 DPO 报错

  #1773 closed Aug 28, 2024

• 关于用自定义agent数据集微调，能不能增加一些自定义数据集的样例

  #1758 closed Aug 28, 2024

• token should only be of type types or str

  #1757 closed Aug 28, 2024

• 可以问一下我们如果自定义数据集的话会是什么样的格式被访问呢 是需要自己封装一下吗

  #1750 closed Aug 28, 2024

• 多工具调用问题

  #1749 closed Aug 28, 2024

• UnicodeDecodeError: 'charmap' codec can't decode byte 0x81 in position 4823: character maps to <undefined>

  #1744 closed Aug 28, 2024

• internvl-8b imblanced GPU allocation with device_max_memory

  #1735 closed Aug 28, 2024

• qwen 2.7B MOE 训练缓慢，GPU 利用率很低。另外，断点续训时，预估的剩余时间需要休修正下

  #1732 closed Aug 28, 2024

• 多机训练InternVL2报错。完全照着官网和赵哥的脚本，多机报错，但单机完全OK

  #1728 closed Aug 28, 2024

• 微调InternVL2特别占显存

  #1726 closed Aug 28, 2024

• 使用webui上传图片会报错

  #1727 closed Aug 28, 2024

• lora微调后推理时遇到：ValueError: Due to limitations of the custom LoRA CUDA kernel, max_num_batched_tokens must be <= 65528 when LoRA is enabled.

  #1723 closed Aug 28, 2024

• args.preprocess_num_proc设置为8报错

  #1718 closed Aug 28, 2024

• Multimodal Document Optimization: Custom Datasets & Inference Documents

  #1716 closed Aug 28, 2024

• qwen2 7B 单机8卡 dpo报错

  #1715 closed Aug 28, 2024

• 纯文本微调GLM-4V报错: ValueError: 151339 is not in list

  #1712 closed Aug 28, 2024

• Update docker image

  #1711 closed Aug 28, 2024

• InternVL2 微调 报错 Message: 'Error occurs in lazy tokenize:'

  #1704 closed Aug 28, 2024

• module 'wandb.proto.wandb_internal_pb2' has no attribute 'Result'

  #1700 closed Aug 28, 2024

• 单机多卡执行 LLaMA-2 全量 SFT 时 NCCL Timeout 报错

  #1690 closed Aug 28, 2024

• 运行swift-webui出错

  #1687 closed Aug 28, 2024

• 模型轻量化问题

  #1683 closed Aug 28, 2024

• 【求助】英伟达 L40 4卡无法微调通义千问2 72B INT4 模型

  #1676 closed Aug 28, 2024

• sft是否支持指定模型的特定参数？

  #1661 closed Aug 28, 2024

• 部署后只能通过ip+端口访问，通过域名访问报错

  #1642 closed Aug 28, 2024

• swift swift-ui error: unable to invoke subcommand: swift-web-ui (No such file or directory)

  #1629 closed Aug 28, 2024

• torch._C._LinAlgError: linalg.cholesky 即使讲quant_n_samples提到了2048依然报错

  #1627 closed Aug 28, 2024

• florence模型微调报错

  #1624 closed Aug 28, 2024

• 你好，想问下swift中template.py文件，怎么样和指令微调一起使用呢？

  #1623 closed Aug 28, 2024

• 不支持--model_type gemma2-2b-instruct

  #1619 closed Aug 28, 2024

• 加载本地下好的sd模型进行微调时，仍然需要去网页端下载模型

  #1616 closed Aug 28, 2024

• GLM4V运行时报错

  #1611 closed Aug 28, 2024

• 请问下GLM4V支持多样本批量请求吗

  #1608 closed Aug 28, 2024

• On the use of --rope_scaling in InternVL2

  #1591 closed Aug 28, 2024

• DPO使用deepspeed出现报错

  #1574 closed Aug 28, 2024

• 微调, deepspeed出现报错

  #1570 closed Aug 28, 2024

• DDP in DPO tuning on MLLM

  #1549 closed Aug 28, 2024

• swift pt命令报错，把缓存的数据集全删了重新下载 还是报错

  #1544 closed Aug 28, 2024

• device_map error for internVL2.0, trained on multi-node

  #1538 closed Aug 28, 2024

• No module named 'swift.cli.main'问题

  #1536 closed Aug 28, 2024

• 微调deepseekcoderv2后推理明显变慢，如何只activate moe的参数？

  #1535 closed Aug 28, 2024

• 微调后模型进行量化出现错误

  #1533 closed Aug 28, 2024

• InternVL2的grounding任务自定义数据集如何一个prompt里融合多个<box>或<ref-object>

  #1532 closed Aug 28, 2024

• Florence-2: How to add custom tokens during fine-tuning training?

  #1524 closed Aug 28, 2024

• 在cogvlm2模型上支持多模态大模型的强化学习训练，例如DPO训练

  #1523 closed Aug 28, 2024

• KTO训练数据集报错

  #1513 closed Aug 28, 2024

• flashattention3

  #1493 closed Aug 28, 2024

• 目前支持并行工具调用嘛？

  #1491 closed Aug 28, 2024

• swift可以训练量化之后的模型吗，比如modelscope里面的awq或者gptq量化之后的模型

  #1472 closed Aug 28, 2024

• intervl 微调后lora合并推理和直接用原来demo推理结果不一致

  #1471 closed Aug 28, 2024

• function call微调数据格式

  #1458 closed Aug 28, 2024

• get_dataset 函数报错，疑似是加载数据太多导致

  #1453 closed Aug 28, 2024

• DPO训练报错 AttributeError: 'Seq2SeqTrainingArguments' object has no attribute 'model_init_kwargs'

  #1434 closed Aug 28, 2024

• Finetuning LLaVA Mistral, gets error when working on multiple GPUs: ValueError: Expected input batch_size (1063) to match target batch_size (2215).

  #1399 closed Aug 28, 2024

• 支持对NPU AWQ量化的模型推理部署吗？

  #1429 closed Aug 28, 2024

• 使用lora进行resume from checkpoint时，无法加载模型

  #1207 closed Aug 28, 2024

• Q-GaLore support

  #1388 closed Aug 28, 2024

• 如何预读取数据

  #1379 closed Aug 28, 2024

• CogVLM2 Video

  #1339 closed Aug 28, 2024

• 不支持DDP多卡推理

  #1377 closed Aug 28, 2024

• 可以支持自定义评测集合与评测指标吗

  #1374 closed Aug 28, 2024

• 脚本使用问题

  #1358 closed Aug 28, 2024

• NPU微调报错AssertionError: Torch not compiled with CUDA enabled ，模型时qwen-7b-chat

  #1353 closed Aug 28, 2024

• agent部署如何实现openai function call的样式，而不是如文档中的react

  #1340 closed Aug 28, 2024

• VLM是否支持并行推理？

  #1330 closed Aug 28, 2024

• 使用多进程处理数据时，报错TypeError: _LazyConfigMapping.__init__() missing 1 required positional argument: 'mapping'

  #1328 closed Aug 28, 2024

• 自定义数据集文档看不太懂

  #1323 closed Aug 28, 2024

• 请问为什么在训练和merge完lora推理测试的时候log没有打印出custom_val_dataset中的问题和回答

  #1322 closed Aug 28, 2024

• 可否缓存预处理后的数据集？

  #1315 closed Aug 28, 2024

• 微调时显存都占得很满

  #1314 closed Aug 28, 2024

• --max_length 参数是否不对自定义数据集生效

  #1297 closed Aug 28, 2024

• 微调模型报错

  #1296 closed Aug 28, 2024

• swift 接入数据集下载后还是会不断请求网络

  #1285 closed Aug 28, 2024

• Quantization stops for no reason when GPTQ quantizing Qwen2-72b-Instruct

  #1121 closed Aug 28, 2024

• MiniCPM-V多卡训练模型infer与单卡不一致

  #1191 closed Aug 28, 2024

• SimPO不支持zero3_offload分布式训练

  #1095 closed Aug 28, 2024

• 无法运行swift

  #1107 closed Aug 28, 2024

• GLM4V微调OOM

  #1122 closed Aug 28, 2024

• 请问有多模态模型可以实现few-shot（in-context）数据的微调吗？

  #1094 closed Aug 28, 2024

• 低成本更新swfit版本

  #1089 closed Aug 28, 2024

• 微调qwen1.5 110B chat后会重复说

  #1084 closed Aug 28, 2024

• 关于VLM的评测支持

  #1117 closed Aug 28, 2024

• failed (exitcode: -9) local_rank: 0 (pid: 1760809) of binary: /home/xxx/miniconda3/envs/swift/bin/python

  #1208 closed Aug 28, 2024

• Qwen-VL-Chat全参精调后无法输出eos_token

  #1215 closed Aug 28, 2024

• Internvl-chat-v1.5使用多卡推理报错

  #1228 closed Aug 28, 2024

• 之前用命令进行chatglm3进行swif训练，能够训练成功现在训练一会就报错内存已经满了 是什么原因

  #1230 closed Aug 28, 2024

• 多模态模型如何多卡部署

  #1236 closed Aug 28, 2024

• Why has the memory usage of the latest code increased? It seems there is now a data loading module.

  #1253 closed Aug 28, 2024

• Why does the latest code seem to have slower training speed and increased memory usage?

  #1255 closed Aug 28, 2024

• The acc is different when infering a glm4v model trained based on the Tutorial

  #1258 closed Aug 28, 2024

• qwen2 系列模型后续是否会支持 fp16 微调？

  #1262 closed Aug 28, 2024

• Finetuning Florence on the forgetting problem

  #1267 closed Aug 28, 2024

• Not working with python virtualenv

  #1266 closed Aug 28, 2024

• How to set different learning rate for different groups of parameters in fine-tuning?

  #1271 closed Aug 28, 2024

• Internvl 微调后的模型，python代码在哪里加载自己模型的路径

  #1281 closed Aug 28, 2024

• swift 能够在k8s中训练推理运行吗

  #1283 closed Aug 28, 2024

• 能否基于lora或llamapro实现增量学习

  #1123 closed Aug 28, 2024

• Model is not saved when resume_from_checkpoint

  #1829 closed Aug 28, 2024

• 使用internlm-xcomposer2_5-7b-chat进行dpo训练，数据报错

  #1831 closed Aug 28, 2024

• minicpm-v 2.5 sft训练时跑验证集会额外占用10GB显存是正常现象吗？

  #1140 closed Aug 28, 2024

• GLM4v LORA微调后，断点训练失败

  #1133 closed Aug 28, 2024

• 请问swift训练MiniCPM-v2.0时应该如何制作带坐标的数据集

  #1137 closed Aug 28, 2024

• 用户如何自己实现新的参数高效微调算法？

  #1147 closed Aug 28, 2024

• 量化训练qwen2-72b-instruct 的权重突入中断

  #1151 closed Aug 28, 2024

• GLM4v使用Lora模型微调后，merge模型后运行报错，

  #1163 closed Aug 28, 2024

• 预训练模板错误

  #1166 closed Aug 28, 2024

• 关于"自我认知微调最佳实践"和"自定义模型和数据集"的说明优化

  #1176 closed Aug 28, 2024

• 训练及导出显存消耗不正常

  #1179 closed Aug 28, 2024

• glm4v微调到一半会显示网络问题，请问有其他加载本地输入的方法吗

  #1185 closed Aug 28, 2024

• 量化后的模型推理报错怎么解决

  #940 closed Aug 28, 2024

• 多节点训练报错

  #945 closed Aug 28, 2024

• 我想将模型保存到本地，怎么才能保存。同时怎么调用本地的模型？

  #950 closed Aug 28, 2024

• 使用zero3进行多机多卡全量微调，保存的模型权重不完整

  #972 closed Aug 28, 2024

• 微调量化后qwen1half-14b-chat-gptq-int8推理时向量报错RuntimeError: probability tensor contains either `inf`, `nan` or element < 0

  #994 closed Aug 28, 2024

• 用qwen-7b-int4和int8进行lora微调后，微调和推理没问题，但部署后，请求报错

  #935 closed Aug 28, 2024

• 训练qwen14b，前面lr一直为0

  #943 closed Aug 28, 2024

• 多机多卡推理

  #927 closed Aug 28, 2024

• 微调minicpmv2时cpu占用率超高

  #1008 closed Aug 28, 2024

• 对qwen2-72b-instruct-awq进行qlora微调之后，如何进行merge?

  #1839 closed Aug 28, 2024

• Infer internlm-xcomposer2 lead to `ValueError: Input length of input_ids is 0, but `max_length` is set to -1066.`

  #862 closed Aug 28, 2024

• react模板放在system和user的区别？

  #1025 closed Aug 28, 2024

• bitsandbytes was compiled without GPU support

  #1031 closed Aug 28, 2024

• swift训练minicpm-v如何设置grounding格式

  #1077 closed Aug 28, 2024

• 如何批量导出MS数据集为swift能加载的格式

  #1041 closed Aug 28, 2024

• swift微调多模态大模型后，比如Intern VL 1.5，可以使用lmdeploy部署吗？

  #1033 closed Aug 28, 2024

• 是否支持idm-vton模型的微调呢？

  #1040 closed Aug 28, 2024

• 随机初始化预训练

  #1045 closed Aug 28, 2024

• 不太明白Lisa为何能够节省显存，希望能够帮忙解答

  #1056 closed Aug 28, 2024

• [Help]请问MiniCPM2.5支持Web端交互推理吗？

  #1060 closed Aug 28, 2024

• 又测了下，您说的最新的代码是指加了if batch[0].get('pixel_values') is not None:这个判断语句吗，在batch>1的情况下，还是会出现keyerror的

  #1061 closed Aug 28, 2024

• 关于多图微调和推理问题

  #1074 closed Aug 28, 2024

• 请教一下 有没有办法扩展c4ai-command-r-plus的上下文长度呢

  #903 closed Aug 28, 2024

• 使用lora微调合并权重加载模型报错

  #825 closed Aug 28, 2024

• Eval 模块使用时 C-Eval 进行测试时，让人不确定日志输出

  #905 closed Aug 28, 2024

• Lora认知微调，微调成功，直接进行推理也正确。但merge之后，再推理，生成乱码，并报错“TypeError: argument 'tokens': 'NoneType' object cannot be converted to 'PyString'”

  #898 closed Aug 28, 2024

• 基于上次提的问题#691，后续改进后似乎依旧不能按微调的情况回复。

  #880 closed Aug 28, 2024

• VRAM requirement for full sft deepseek VL 7B

  #860 closed Aug 27, 2024

• llama3-8b-instruct awq量化oom

  #852 closed Aug 27, 2024

• 使用DDP运行时显存不够，但是使用Model Parallel时可以正常finetune，耗时很大

  #851 closed Aug 27, 2024

• 请教一下，多节点多卡微调，用Slurm怎样运行？

  #828 closed Aug 27, 2024

• win10训练qwen1.5-moe-A2.7B-chat-gptq-int4速度缓慢

  #811 closed Aug 27, 2024

• Langchain-Chatchat部署训练后的模型后推理异常

  #794 closed Aug 27, 2024

• Support QLoRA with HQQ quantization

  #786 closed Aug 27, 2024

• 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？

  #774 closed Aug 27, 2024

• ddp_timeout能否作为sft命令行参数

  #773 closed Aug 27, 2024

• 希望可以支持自我奖励的优化方法

  #757 closed Aug 27, 2024

• swift 1.8. -> 2.0.0 原lora脚本报错

  #713 closed Aug 27, 2024

• tuner_backend选swift的时候，rsLoRA训练中loss正常下降，但训练后模型推理输出空值

  #707 closed Aug 27, 2024

• 72B的模型首字延时如何减少

  #689 closed Aug 27, 2024

• 部署Qwen1.5-MoE-A2.7B-Chat进行推理问题

  #678 closed Aug 27, 2024

• 期望支持PiSSA微调

  #665 closed Aug 27, 2024

• 自定义推理文件无法使用两卡以上多卡推理

  #638 closed Aug 27, 2024

• windows 系统训练失败，数据集字符编码问题

  #686 closed Aug 27, 2024

• cogvlm推理无法选择大于1的num_beam

  #636 closed Aug 27, 2024

• lora微调cogvlm打开量化报错

  #635 closed Aug 27, 2024

• Unable to enable web UI

  #1756 closed Aug 27, 2024

• lora_modules_to_save + deepspeed zero2 导致无法保存checkpoint

  #628 closed Aug 27, 2024

• 微调Qwen1___5-7B-Chat，4比特量化，为何还需要占用这么多显存？

  #616 closed Aug 27, 2024

• Live inference of fine-tuned multimodal LLMs?

  #615 closed Aug 27, 2024

• 希望能实现训练和推理的时候开启RoPE外推

  #610 closed Aug 27, 2024

• 微调任务是，中断网页再次打开，找回运行时任务，运行状态无法可视化

  #588 closed Aug 27, 2024

• 断点续训时，显存要求增大了？

  #570 closed Aug 27, 2024

• agent微调数据自制，header信息如何添加？

  #544 closed Aug 27, 2024

• 请问Qwen-audio的训练速度，阿里官方达到多少？

  #511 closed Aug 27, 2024

• 请问能给一点实现动态batch size的建议吗？

  #489 closed Aug 27, 2024

• 可以支持一下SPIN自我博弈微调的方法不

  #466 closed Aug 27, 2024

• 在网页版的demo中推理仅仅只能进行语言模型的对话，没有多模态的推理

  #461 closed Aug 27, 2024

• not supported python3.8

  #459 closed Aug 27, 2024

• LongLoRA微调llama2-7b报错

  #449 closed Aug 27, 2024

• 目前swif的lora微调和adapter微调，token embedding方式相同吗

  #391 closed Aug 27, 2024

• web-ui - LLM训练的 system字段 会导致命令注入

  #363 closed Aug 27, 2024

• 使用vllm官方镜像部署swift finetune后的checkpoint 报错

  #1820 closed Aug 27, 2024

• 单机多卡 DDP+MP 断点续训(resume_from_checkpoint)报错

  #596 closed Aug 26, 2024

• How to choose a template (or template_type)

  #1813 closed Aug 25, 2024

• 单卡训练qwen会卡一下

  #1802 closed Aug 23, 2024

• 要如何指定InternLM-XComposer2d5的hd_num参数

  #1799 closed Aug 23, 2024

• About the parameter sft_type

  #1803 closed Aug 23, 2024

• How to load a model from DPO checkpoint

  #1786 closed Aug 22, 2024

• InternVL2-8b DPO Failed mapping

  #1768 closed Aug 22, 2024

• AttributeError: module 'inspect' has no attribute 'getargspec'. Did you mean: 'getargs'?

  #1387 closed Aug 21, 2024

• The deployment of llama3.1-405b-instruct

  #1484 closed Aug 21, 2024

• 评测Internvl76B时，完成了部署但是没有推理

  #1486 closed Aug 21, 2024

• 用改框架微调GLM-4V-9B模型，存在重复输出和任务无法拟合的问题。

  #1502 closed Aug 21, 2024

• GLM 4V 9B RuntimeError: Expected all tensors to be on the same device, but found at least two devices

  #1547 closed Aug 21, 2024

• 全量微调full断点续训(resume_from_checkpoint)问题

  #1577 closed Aug 21, 2024

• sft 增大显卡数，速度没有提升，修改了gradient_accumulation_steps, remaining_time还是不变

  #1674 closed Aug 21, 2024

• The loss is always 0 when fine-tuning internlm-xcomposer2d5-7b

  #1680 closed Aug 21, 2024

• How to enable visual module fine-tuning?

  #1685 closed Aug 21, 2024

• qwen2-audio analysis为啥60s音频文件识别不出文字啊

  #1691 closed Aug 21, 2024

• internlm-xcomposer2-vl-1_8b 训练支持

  #1739 closed Aug 21, 2024

• qwen2-audio only trains the LLM part. How to modify target_regex?

  #1747 closed Aug 21, 2024

• Support using lmdeploy for inference and deployment on videos (internvl2)

  #1679 closed Aug 21, 2024

• 微调InternVL2特别占显存

  #1725 closed Aug 21, 2024

• Error during Finetune MiniCPM 微调MiniCPM时出错 & zero3

  #1135 closed Aug 20, 2024

80 Issues opened by 66 people

• V100显卡 Ubuntu22.04系统 qwen2-vl-2b模型， 单卡测试脚本运行正常，双卡，三卡，四卡运行异常。

  #2087 opened Sep 20, 2024

• Qwen2-VL 72B API推理错误

  #2086 opened Sep 20, 2024

• 昇腾910B单机多卡断点续训问题

  #2085 opened Sep 20, 2024

• npu 推理和部署怎么设置多卡

  #2084 opened Sep 20, 2024

• swift升级到2.4.0及以上版本后加载训练数据集报错

  #2083 opened Sep 20, 2024

• dpo InternVL2-8B meets OOM

  #2082 opened Sep 20, 2024

• I tried to fine tune the instructions for qwen2-7B instruction, qwen2.5-7B instruction, and llama3 7B instruction using the latest version of Swift, but the loss decreased to 2.1~2.2 and could not be reduced further.

  #2079 opened Sep 20, 2024

• 希望支持训练user部分或者同时训练user和assistant部分

  #2078 opened Sep 20, 2024

• MLLM支持KTO训练

  #2076 opened Sep 20, 2024

• Training llama 3.1 70B using 4 A6000

  #2074 opened Sep 19, 2024

• internvl2+lmdeploy部署推理视频，有没有控制采样频率，前处理的参数？目前的推理时间过长了

  #2069 opened Sep 19, 2024

• Fine-tuning best practices for qwen2.5-72b-instruct and qwen2-vl-72b-instruct.

  #2064 opened Sep 18, 2024

• swift更新到最新版后无法使用多个节点训练

  #2057 opened Sep 18, 2024

• 数据预处理报错

  #2055 opened Sep 18, 2024

• Pixtral 12B

  #2053 opened Sep 17, 2024

• 请问如何freeze一部分pretrain的模型后，接入自定义的pytorch model并进行训练？

  #2049 opened Sep 16, 2024

• AttributeError: 'Qwen2VLForConditionalGeneration' object has no attribute 'quantize'. Did you mean: 'dequantize'?

  #2039 opened Sep 14, 2024

• 使用OpenAI API方式启动部署报错：ImportError: cannot import name 'prepare_vllm_engine_template' from 'swift.llm.utils'

  #2038 opened Sep 14, 2024

• LISA训练要么OOM，要么使用deepseed就报错

  #2035 opened Sep 14, 2024

• 关于 微调+量化 的一些经验求教

  #2033 opened Sep 13, 2024

• internvl2-8b在docker里训练OOM

  #2027 opened Sep 12, 2024

• How to track pooling stride and frame count of llava-next-video in Swift

  #2024 opened Sep 12, 2024

• swift不能做模型的二次预训练？具体怎么做啊？

  #2022 opened Sep 12, 2024

• Dataset format compatibility between LLaVA-NexT and Qwen-VL2 for custom JSON datasets

  #2017 opened Sep 11, 2024

• internvl2-40b infer

  #2016 opened Sep 11, 2024

• Deployment or Export

  #2014 opened Sep 11, 2024

• Qwen2-VL-7B-Instruct训练爆显存

  #2010 opened Sep 11, 2024

• Regular lora target module cannot use imdeploy

  #2007 opened Sep 10, 2024

• train_on_input

  #2004 opened Sep 10, 2024

• inference接口不支持设置返回多组结果

  #2002 opened Sep 10, 2024

• 无法评测LoRA微调后的llava1.5模型

  #2000 opened Sep 10, 2024

• MiniCPM-V-2 lora微调后推理报错： AssertionError: Current sentence length exceeds the model max_length: 4096

  #1999 opened Sep 10, 2024

• internlm2_5-20b-chat 量化模型不支持vllm推理

  #1998 opened Sep 10, 2024

• internvl 40Bckp保存和推理加载的时候出错

  #1993 opened Sep 9, 2024

• Inference Not Working on cuda:1

  #1991 opened Sep 9, 2024

• LLaVA-NeXT-Video model configuration initialize error

  #1986 opened Sep 9, 2024

• llava1_5-7b-instruct 评测开始时能正常输出，后面无法正常输出

  #1978 opened Sep 8, 2024

• Dataset preparation for Object Detection with Florence2

  #1977 opened Sep 8, 2024

• mplug-owl3-7b-chat fine-tuning document

  #1969 opened Sep 7, 2024

• Dataset stucked when using --dataloader_num_workers 1 and --streaming true

  #1968 opened Sep 7, 2024

• 全量微调minicpm-V2.6,但是生成的sft_args.json里面仍然有lora

  #1967 opened Sep 6, 2024

• 如果我在训练Lora时也想训练head，应该如何设置

  #1965 opened Sep 6, 2024

• 用deploy部署qwen2vl，多个请求同时并发报错

  #1961 opened Sep 6, 2024

• 期望RLHF能支持序列并行（sequence_parallel）

  #1958 opened Sep 6, 2024

• qwen2-vl-7b-instruct 以VLLM形式启动推理引擎失败“ assert "factor" in rope_scaling”

  #1954 opened Sep 5, 2024

• swift 中如何设置模型内部参数，如internvl中的max_dynamic_patch

  #1953 opened Sep 5, 2024

• NPU qwen2模型推理报错

  #1951 opened Sep 5, 2024

• 请问是否支持在NPU上训练多模态大模型

  #1941 opened Sep 5, 2024

• 微调速度非常慢

  #1940 opened Sep 5, 2024

• streaming模式读取数据，显存利用率很低

  #1939 opened Sep 5, 2024

• Training stops for `KTO` after model loads into memory.

  #1938 opened Sep 4, 2024

• qwen2_audio_7b_instruct利用VLLM推理错误

  #1937 opened Sep 4, 2024

• 关于 Qwen2_VL-2B 微调时显存不足的问题

  #1935 opened Sep 4, 2024

• 微调glm4v, 给glm4v的视觉部分都添加了checkpoint, 但是还是显存溢出（lora_target_modules 设置为'ALL' ）

  #1934 opened Sep 4, 2024

• “训练推理界面”后点击“通过 API 使用”时报错。

  #1932 opened Sep 4, 2024

• 视觉模块支持gradient_checkpointing

  #1928 opened Sep 4, 2024

• minicpm-v-v2.6评测出现结果为0情况

  #1924 opened Sep 3, 2024

• Support for Fine-Tuning Best Practices with LLaVA-OV

  #1923 opened Sep 3, 2024

• 【新增功能需求】 Internvl2模型+VLLM 后端实现 异步客户端请求的Video 推理功能

  #1921 opened Sep 3, 2024

• MooER audio support request

  #1919 opened Sep 3, 2024

• llava-llama-3-8b-v1_1 AttributeError: 'NoneType' object has no attribute 'get_output_embeddings'

  #1911 opened Sep 3, 2024

• qwen2-vl-chat-instruct示例数据格式

  #1896 opened Sep 2, 2024

• internvl2-llama3-76b 微调报错

  #1892 opened Sep 2, 2024

• 如何调整训练损失函数类型？自定义损失函数？(Custom loss function)

  #1891 opened Sep 2, 2024

• qwen2-vl微调使用flash_attn报错

  #1887 opened Sep 2, 2024

• CUDA error: too many resources requested for launch (V100, qwen2-vl)

  #1867 opened Aug 30, 2024

• Memory(not GPU RAM) exceeds when using 'swift deploy'

  #1866 opened Aug 30, 2024

• minicpm-V-2最佳实践,执行推理时,模型不输出任何结果

  #1863 opened Aug 30, 2024

• 🎉Support for finetuning of Qwen2-VL-Chat series models

  #1857 opened Aug 29, 2024

• 希望对vllm的cpu运行模式增加支持

  #1844 opened Aug 29, 2024

• How to use XLA/TPU? Potentially SPMD FSDP with InternVL model

  #1837 opened Aug 27, 2024

• 多机训练速度问题

  #1825 opened Aug 27, 2024

• 训练中途突然报错 NCCL watchdog thread terminated with exception

  #1817 opened Aug 26, 2024

• Phi3.5-vision-instruct fine-tuning best practices. (Latex OCR Fine-tuning)

  #1809 opened Aug 23, 2024

• 增加推理加速器适配

  #1808 opened Aug 23, 2024

• lr_scheduler_type

  #1807 opened Aug 23, 2024

• [BUG] chatGLM4微调，设置predict_with_generate=True, 评估时间过长

  #1792 opened Aug 22, 2024

• 多模态评测数据下载失败

  #1787 opened Aug 22, 2024

• 使用Lora微调MiniCPM-V-2_6，合并后再Lora训练出现问题

  #1778 opened Aug 21, 2024

• RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

  #1777 opened Aug 21, 2024

28 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.

• 书生26B合并dpo后的适配器失败

  #1667 commented on Aug 23, 2024 • 0 new comments

• 能否支持/v1/embeddings的api调用

  #807 commented on Aug 27, 2024 • 0 new comments

• export problem: get_model_tokenizer_with_flash_attn() got multiple values for keyword argument 'automodel_class'

  #836 commented on Aug 28, 2024 • 0 new comments

• 2.0.4之后的版本的显存使用问题

  #922 commented on Aug 28, 2024 • 0 new comments

• 是否支持自定义lr_scheduler

  #1075 commented on Aug 28, 2024 • 0 new comments

• Loss and acc drop to 0 after several steps

  #1062 commented on Aug 28, 2024 • 0 new comments

• 可以支持一下InternLM2-Math-Plus-Mixtral8x22B的微调吗

  #1019 commented on Aug 28, 2024 • 0 new comments

• 希望能应用TensorRT加速训练和推理

  #942 commented on Aug 28, 2024 • 0 new comments

• Process hang with futex(0x7f403c0199d0, FUTEX_WAIT, 14826, NULL

  #1128 commented on Aug 28, 2024 • 0 new comments

• minicpm-v-v2_5-chat 微调vpm显存溢出

  #1286 commented on Aug 28, 2024 • 0 new comments

• 自定义评测数据集做评测时出现，模型用vllm.entrypoints.openai.api_server起的。运行评测脚本出现错误

  #1295 commented on Aug 28, 2024 • 0 new comments

• 关于agent微调数据问题

  #1351 commented on Aug 28, 2024 • 0 new comments

• Representing results of Agent best practice with Qwen2-7b-instruct outputs unexpected <|endoftext|> and <|im_start|>

  #1155 commented on Aug 28, 2024 • 0 new comments

• Florence2 batched inference

  #1441 commented on Aug 28, 2024 • 0 new comments

• 能否加上昇腾NPU上多卡推理

  #1469 commented on Aug 28, 2024 • 0 new comments

• smooth quant support

  #1489 commented on Aug 28, 2024 • 0 new comments

• RAG支持

  #1548 commented on Aug 28, 2024 • 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 Aug 28, 2024 • 0 new comments

• 是否考虑支持RLHV-V中提出的DDPO

  #1639 commented on Aug 28, 2024 • 0 new comments

• support llama3 megatron

  #1736 commented on Aug 28, 2024 • 0 new comments

• 单机三卡微调50K数据OOM

  #1729 commented on Aug 30, 2024 • 0 new comments

• swift 量化多模态大模型internvl2-26B，报错

  #1504 commented on Aug 31, 2024 • 0 new comments

• 魔搭NPU训练部署交流群

  #1589 commented on Sep 4, 2024 • 0 new comments

• AttributeError: module 'transformers_modules.InternVL2-2B-1epoch.tokenization_internlm2' has no attribute 'InternLM2Tokenizer'

  #1663 commented on Sep 8, 2024 • 0 new comments

• SWIFT 2.4 TO DO LIST

  #1617 commented on Sep 10, 2024 • 0 new comments

• Best practice for Qwen2-Audio

  #1653 commented on Sep 11, 2024 • 0 new comments

• Best Practices for Inference and Fine-Tuning with MiniCPM-V 2.6

  #1613 commented on Sep 15, 2024 • 0 new comments

• DPO训练的时候grad_norm出现nan值

  #923 commented on Sep 17, 2024 • 0 new comments