v0.4.1

What's Changed

• 【开源实习】Speech2Text模型迁移 by @imeet000 in #1725
• 【开源实习】Unispeech模型迁移 by @liuyifan123123 in #1722
• Fix typing mistakes in documentation by @Yashbhatt786 in #1740
• support O2 on OrangePi, tinyllama 450ms -> 160ms by @lvyufeng in #1744
• fix no_grad state error by @lvyufeng in #1754
• avoid lack of cumsum operator on OrangePi by @lvyufeng in #1755
• fix enable_grad by @lvyufeng in #1757
• fix name for nn.Linear by @lvyufeng in #1758
• fix nll_loss casued backward error by @lvyufeng in #1759
• fix out_channels of Linear by @lvyufeng in #1760
• add toc in readme.md by @adarsh-jha-dev in #1746
• safe_load_file use mmap to speedup by @lvyufeng in #1763
• fix from_numpy caused error by @lvyufeng in #1765
• fix llama embedding by @lvyufeng in #1774
• upgrade wav2vec by @lvyufeng in #1776
• update nn.utils.parametrizations by @lvyufeng in #1777
• fix sew and sew_d by @lvyufeng in #1782
• fix whisper ut by @lvyufeng in #1786
• fix llava on Ascend by @lvyufeng in #1788
• fix mixtral ut by @lvyufeng in #1790
• fix gamma on CPU and GPU by @lvyufeng in #1791
• 【开源之夏】add dpo trainer and support dpo training for mindnlp by @Trace2333 in #1701
• 【开源实习】GPT-J-6B模型应用开发 by @LuMH1027 in #1742
• 【开源实习】ImageGPT模型应用开发 by @Helloyouth2022 in #1700
• 【开源实习】X_CLIP模型应用开发 by @dyedd in #1694
• 【开源实习】UPerNet模型应用开发 by @Maysixi in #1717
• update whisper stream by @lvyufeng in #1796
• 【开源实习】multilayer_perceptron_lora模型微调 by @barryyfli in #1749
• dino应用开发 by @LuMH1027 in #1797
• support depth_estimation & doc_qa by @lvyufeng in #1800
• support fill_mask/image_classification/image_feature_extraction pipelines by @lvyufeng in #1801
• init mimm & move tests folder by @lvyufeng in #1804
• fix per_gpu args by @lvyufeng in #1809
• add mindspore infer function patch by @lvyufeng in #1810
• fix sync parallel and support low_cpu_mem_usage by @lvyufeng in #1814
• 【开源实习】dpt模型应用开发 by @yegoling in #1813
• 【开源实习】LayoutLMv2模型应用开发 by @Helloyouth2022 in #1805
• 【开源实习】BERT模型应用开发 by @KhunLounZai in #1781
• 【开源实习】peft_adalora_seq2seq模型微调 by @liuyifan123123 in #1747
• 【开源实习】Vision Transformer模型应用开发 by @KhunLounZai in #1643
• 【开源实习】SegFormer模型应用开发 by @KhunLounZai in #1663
• 【开源实习】BEiT模型应用开发 by @somecatw in #1680
• 【开源实习】SAM模型应用开发 by @KhunLounZai in #1688
• Update prompt_direct.txt by @YadlaMani in #1732
• fix:解决minicpm未注册问题 by @ResDream in #1821
• Module support H2D move by @lvyufeng in #1831
• fix low_cpu_mem_usage(contguous) by @lvyufeng in #1832
• value_and_grad support attacach grads, Parameter support accumulate a… by @lvyufeng in #1833
• fix update_and_allocate by @lvyufeng in #1834
• value_and_grad support kwargs by @lvyufeng in #1835
• 修复blip推理报错 by @confused666 in #1838
• fix roll on CPU by @lvyufeng in #1840
• deprecated mindnlp.transformers.optimization by @lvyufeng in #1841
• fix optimizer args as same dtype on GPU by @lvyufeng in #1842
• image_classification_timm_peft_lora模型微调 by @chenxinxi in #1830
• 【开源实习】 MobileViTV2 模型迁移 by @oucfm in #1850
• 重做deepseek_v2模型并补注册 by @ShangJingLi in #1859
• fix: Fix problem using ops.ones in BigBird by @reeered in #1861
• #benchmark: add GLUE-QNLI benchmark, including 10 models inference accuracy comparsion by @xuhangscut in #1865
• #fix benchmark GLUE-QNLI fix read_csv error and predict funciton and modify readme description by @xuhangscut in #1868
• feat: add data parallel of native mindspore to mindnlp.Trainer.base by @Tridu33 in #1852
• Revert "image_classification_timm_peft_lora模型微调" by @lvyufeng in #1871
• add minicpm3 model and dynmaic inference demo by @xing-yiren in #1870
• fix baichuan finfo error by @lvyufeng in #1872
• fix sbert precision problem on mindnlp.sentence by @lvyufeng in #1873
• fix cell_ to module_ in mindnlp.peft by @lvyufeng in #1874
• fix cells to modules in mindnlp.peft by @lvyufeng in #1875
• fix named_modules by @lvyufeng in #1877
• fix sbert normalize_embeddings by @lvyufeng in #1879
• 【开源实习】MindSpore自定义RWKV算子开发（Python接口实现） by @EliwiiKeeya in #1862
• fix llama and baichuan typo by @lvyufeng in #1883
• update core.ops with pyboost by @lvyufeng in #1884
• fix pytest error by @lvyufeng in #1885
• fix mobilebert register by @lvyufeng in #1890
• feat: add file lock for remote files download to local path when multiple thread environment. by @Tridu33 in #1887
• fix ia3 by @lvyufeng in #1891
• expose prob as positional argument for bernoulli ops by @qhzhuang in #1949
• fix model llama for split function in line 852 by @zhuizhuzheming in #1941
• fix TensorPy for mindspore 2.5 by @lvyufeng in #1961
• fix TensorPy init empty Tensor by @lvyufeng in #1962
• fix Blip2加载和推理bug #1902 #1904 #1905 by @Alemax067 in #1958
• 【开源实习】Mask2Former模型应用开发 by @linrx-ctrl in #1770
• 【开源实习】TAPAS模型应用开发 by @sjtu-weimang in #1839
• 【开源实习】MaskFormer模型应用开发 by @linrx-ctrl in #1864
• 【开源实习】YOLOS模型应用开发 by @chenxinxi in #1867
• 【开源实习】bert_japanese模型微调 by @zhuyuhua1 in #1892
• 【开源实习】bloom模型微调 by @guyueyuan in #1907
• 【开源实习】bert_generation模型微调 by @guyueyuan in #1911
• 【开源实习】bertweet模型微调 by @Alemax067 in #1964
• 【开源实习】blip_2模型微调 by @Alemax067 in #1965
• ViTMAE模型应用开发 by @1hb6s7t in #1956
• fix mindspore2.5-2.6 caused error by @lvyufeng in #1985
• load safetensor back to numpy by @lvyufeng in #1986
• support triton self-defined op by @lvyufeng in #1990
• 开源实习 BEiT 模型微调 by @4everImmortality in #1975
• 【开源实习】blenderbot模型微调 by @ZhFuGui in #1978
• 【开源实习】blenderbot_small模型微调 by @outbreak-sen in #1980
• fix bugs for ms2.5 by @lvyufeng in #1991
• Fix incorrect import quick_start.md by @XueyanZh...

v0.4.0

时隔5个月，MindNLP 0.4来啦！！！这一次在开源社区同学们的努力下带来了数个重大特性更新！！！

1. 模型大量更新，支持250+ transformers模型，覆盖90% huggingface模型，使用MindSpore进行模型复现、开发训练再无迁移之忧（社区会同步huggingface和业界新模型，随时更新）。
2. 全平台支持：全面支持昇腾910A、910B、310B(香橙派)、GPU、CPU。（PS：目前香橙派唯一可用的AI开发套件）
3. 分布式并行推理：支持多卡多进程并行推理（超过10B的模型可以玩啦！）
4. 量化算法支持：香橙派支持Smooth Quant，GPU支持bitsandbytes int8量化
5. Sentence transformer支持：可以开心的搞RAG啦
6. 动态图性能优化：昇腾硬件下动态图推理速度达到Pytorch+GPU同水平（Llama实测性能85ms/token）
7. 真正的动静统一：通过mindspore.jit一行切换图模式，完全使用huggingface code style，既保证易用性，又能快速提升性能。昇腾硬件下Llama实测性能为动态图2倍(45ms/token)，与mindspore其他静态图based套件性能完全一致。
8. 海量LLM应用更新：包括文本信息抽取、聊天机器人、语音识别、ChatPDF、音乐生成、代码生成、声音克隆等等，伴随着模型支持度的提高，更多好玩的应用等你来开发！

欢迎大家使用、提Issue和Star一下~

安装方法：

pip install mindnlp

代码仓：
https://github.com/mindspore-lab/mindnlp

PS：获取昇腾算力的几个途径：

• 参加MindSpore社区实习和社区活动，免费领取华为云代金券：https://gitee.com/mindspore/community/issues/IAUJE8
• 使用启智社区免费算力：https://www.openi.org.cn/

v0.3.1

What's Changed

• add musicgen gradio demo by @lvyufeng in #1066
• Peft lokr support by @ultranationalism in #1067
• use mkdocs to generate documentation by @lvyufeng in #1070
• Add tutorial quick start by @hubosynapse in #1069
• fix precision error of flash_attn kernel in some case by @hypertseng in #1072
• support prompt_tuning by @lvyufeng in #1073
• add peft integration by @lvyufeng in #1076
• implement cogvlm model and demo by @antigone660 in #1075
• fix pipeline base error by @lvyufeng in #1077
• peft module documentation completion by @ultranationalism in #1078
• Add MindNLP Starcoder vscode demo by @xing-yiren in #1079
• fix starcoder example by @lvyufeng in #1080
• update docs by @lvyufeng in #1084

New Contributors

• @ultranationalism made their first contribution in #1067
• @antigone660 made their first contribution in #1075
• @xing-yiren made their first contribution in #1079

Full Changelog: v0.3.0...v0.3.1

v0.3.0

What's Changed

• add new Trainer like hf-transformers
• Add timesformer, MPNet, pegasus, segformer, sam, llama3, MPT,
  olmo, llava_next, llava, vipllava, convnext, cvt, resnet, van, openelm, phi3
• support flash_attn_v2, flash_attn_bwd
• peft module support ia3, adalora
• support resume download
• add gsm8k cot prompt example

New Contributors

• @Brain100destruction made their first contribution in #1038
• @wjy4399 made their first contribution in #1042
• @zhouyifeng888 made their first contribution in #1047

Full Changelog: v0.2.4...v0.3.0

v0.2.4

What's Changed

• update flashattention kernel by @hypertseng in #962
• support for internlm by @NANDSS1 in #961
• add qwen2_moe & fix bugs by @lvyufeng in #965
• skip big and slow ut on CPU by @lvyufeng in #966
• Add table_transformer model by @ShaoqLin in #976
• fix find_cuda_home raise error by @lvyufeng in #967
• fix error on macOS by @lvyufeng in #977
• add jamba by @lvyufeng in #978
• add bit by @lvyufeng in #979
• add musicgen_melody by @lvyufeng in #980
• Fix tokenized index problem by @WilliamLiuAtCPC in #982
• add text2vec module by @lvyufeng in #983
• support blenderbot by @lvyufeng in #984
• support blenderbot samll by @lvyufeng in #985
• support blip by @lvyufeng in #987
• Add ConvBert model and pass ut. by @neoming in #991
• add model: xlnet by @2802427218 in #988
• support blip_2 by @lvyufeng in #993
• support bridgetower & bros model by @lvyufeng in #994
• Added hypercomplex Tensor Decomposition and example of using it by @maxmgrdv in #995
• loar fintune convbert by @neoming in #997
• support jetmoe & fix python id() caused bugs by @lvyufeng in #998
• update supported list by @lvyufeng in #999

New Contributors

• @NANDSS1 made their first contribution in #961
• @ShaoqLin made their first contribution in #976
• @WilliamLiuAtCPC made their first contribution in #982

Full Changelog: v0.2.3...v0.2.4

v0.2.3

fix layoutlm & use hf-mirror for all urls (#963)

v0.2.2

New models

• ALIGN
• AltCLIP
• Audio Spectrogram Transformer
• BioGPT

What's Changed

• add tokenization_layoutlm by @FLoutione in #935
• support wav2vec2_with_lm by @lvyufeng in #936
• fix whisper inference on Ascend by @lvyufeng in #937
• fix whisper return_token_timestamps error by @lvyufeng in #938
• add evaluate module by @lvyufeng in #939
• add biogpt by @lvyufeng in #940
• support align model by @lvyufeng in #942
• fix _scaled_dot_product_attention bug by @hypertseng in #941
• add altclip & ast model by @lvyufeng in #944

Full Changelog: v0.2.1...v0.2.2

v0.2.1

New features

Pipelines support

• Automatic Speech Recognition
• Question Answering
• Text to Text
• Text Classification
• Text Generation

What's Changed

• update readme by @lvyufeng in #889
• correct bark encodec inference bug by @JinPan-Second in #892
• support starcoder2 by @lvyufeng in #898
• big_bird has a ut failure by @stringency in #905
• fix baichuan error by @lvyufeng in #909
• fix big_bird error by @lvyufeng in #910
• support mixtral by @lvyufeng in #912
• windows use synchronize since fatal errors by @lvyufeng in #914
• support mamba by @lvyufeng in #916
• mamba support graph mode by @lvyufeng in #919
• Text generation by @2802427218 in #918
• Text2Text generation pipeline feat by @fanxing-6 in #917
• fix codegen error on GPU by @lvyufeng in #920
• Added dual-valued Bert model and example of using it by @maxmgrdv in #921
• fix hypercomplex casued errors by @lvyufeng in #922
• add:question answering by @2802427218 in #923
• fix mixtral ut on windows by @lvyufeng in #924
• layoutlm module by @FLoutione in #925
• update peft examples by @lvyufeng in #926
• udate ci pipeline by @lvyufeng in #927
• support deberta model by @lvyufeng in #928
• add bce example by @lvyufeng in #930
• support reformer by @lvyufeng in #931
• fix roll not support on CPU by @lvyufeng in #932
• support automatic_speech_recognition pipeline by @lvyufeng in #934

New Contributors

• @stringency made their first contribution in #905
• @2802427218 made their first contribution in #918
• @fanxing-6 made their first contribution in #917
• @maxmgrdv made their first contribution in #921

Full Changelog: v0.2.0...v0.2.1

v0.2.0

only download models from huggingface (#888)

v0.1.1

Correct checkpoint`s URL for OPT Model. (#648)