[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-10719":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":16,"stars7d":16,"stars30d":17,"stars90d":16,"forks30d":16,"starsTrendScore":16,"compositeScore":18,"rankGlobal":10,"rankLanguage":10,"license":19,"archived":20,"fork":20,"defaultBranch":21,"hasWiki":20,"hasPages":20,"topics":22,"createdAt":10,"pushedAt":10,"updatedAt":32,"readmeContent":33,"aiSummary":34,"trendingCount":16,"starSnapshotCount":16,"syncStatus":17,"lastSyncTime":35,"discoverSource":36},10719,"EasyLM","young-geng\u002FEasyLM","young-geng","Large language models (LLMs) made easy, EasyLM is a one stop solution for pre-training, finetuning, evaluating and serving LLMs in JAX\u002FFlax.","",null,"Python",2515,260,37,28,0,2,59.45,"Apache License 2.0",false,"main",[23,24,25,26,27,28,29,30,31],"chatbot","deep-learning","flax","jax","language-model","large-language-models","llama","natural-language-processing","transformer","2026-06-12 04:00:52","# EasyLM\nLarge language models (LLMs) made easy, EasyLM is a one stop solution for\npre-training, finetuning, evaluating and serving LLMs in JAX\u002FFlax. EasyLM can\nscale up LLM training to hundreds of TPU\u002FGPU accelerators by leveraging\nJAX's pjit functionality.\n\n\nBuilding on top of Hugginface's [transformers](https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Ftransformers\u002Fmain\u002Fen\u002Findex)\nand [datasets](https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Fdatasets\u002Findex), this repo provides\nan easy to use and easy to customize codebase for training large language models\nwithout the complexity in many other frameworks.\n\n\nEasyLM is built with JAX\u002FFlax. By leveraging JAX's pjit utility, EasyLM is able\nto train large models that don't fit on a single accelerator by sharding\nthe model weights and training data across multiple accelerators. Currently,\nEasyLM supports multiple TPU\u002FGPU training in a single host as well as multi-host\ntraining on Google Cloud TPU Pods.\n\nCurrently, the following models are supported:\n* [LLaMA](https:\u002F\u002Farxiv.org\u002Fabs\u002F2302.13971)\n* [LLaMA 2](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.09288)\n* [LLaMA 3](https:\u002F\u002Fllama.meta.com\u002Fllama3\u002F)\n\n## Discord Server\nWe are running an unofficial Discord community (unaffiliated with Google) for discussion related to training LLMs in JAX. [Follow this link to join the Discord server](https:\u002F\u002Fdiscord.gg\u002FRf4drG3Bhp). We have dedicated channels for several JAX based LLM frameworks, include EasyLM, [JaxSeq](https:\u002F\u002Fgithub.com\u002FSea-Snell\u002FJAXSeq), [Alpa](https:\u002F\u002Fgithub.com\u002Falpa-projects\u002Falpa) and [Levanter](https:\u002F\u002Fgithub.com\u002Fstanford-crfm\u002Flevanter).\n\n\n## Models Trained with EasyLM\n### OpenLLaMA\nOpenLLaMA is our permissively licensed reproduction of LLaMA which can be used\nfor commercial purposes. Check out the [project main page here](https:\u002F\u002Fgithub.com\u002Fopenlm-research\u002Fopen_llama).\nThe OpenLLaMA can serve as drop in replacement for the LLaMA weights in EasyLM.\nPlease refer to the [LLaMA documentation](docs\u002Fllama.md) for more details.\n\n\n### Koala\nKoala is our new chatbot fine-tuned on top of LLaMA. If you are interested in\nour Koala chatbot, you can check out the [blogpost](https:\u002F\u002Fbair.berkeley.edu\u002Fblog\u002F2023\u002F04\u002F03\u002Fkoala\u002F)\nand [documentation for running it locally](docs\u002Fkoala.md).\n\n\n## Installation\nThe installation method differs between GPU hosts and Cloud TPU hosts. The first\nstep is to pull from GitHub.\n\n``` shell\ngit clone https:\u002F\u002Fgithub.com\u002Fyoung-geng\u002FEasyLM.git\ncd EasyLM\nexport PYTHONPATH=\"${PWD}:$PYTHONPATH\"\n```\n\n#### Installing on GPU Host\nThe GPU environment can be installed via [Anaconda](https:\u002F\u002Fwww.anaconda.com\u002Fproducts\u002Fdistribution).\n\n``` shell\nconda env create -f scripts\u002Fgpu_environment.yml\nconda activate EasyLM\n```\n\n#### Installing on Cloud TPU Host\nThe TPU host VM comes with Python and PIP pre-installed. Simply run the following\nscript to set up the TPU host.\n\n``` shell\n.\u002Fscripts\u002Ftpu_vm_setup.sh\n```\n\n\n## [Documentations](docs\u002FREADME.md)\nThe EasyLM documentations can be found in the [docs](docs\u002F) directory.\n\n\n## Reference\nIf you found EasyLM useful in your research or applications, please cite using the following BibTeX:\n```\n@software{geng2023easylm,\n  author = {Geng, Xinyang},\n  title = {EasyLM: A Simple And Scalable Training Framework for Large Language Models},\n  month = March,\n  year = 2023,\n  url = {https:\u002F\u002Fgithub.com\u002Fyoung-geng\u002FEasyLM}\n}\n```\n\n\n\n## Credits\n* The LLaMA implementation is from [JAX_llama](https:\u002F\u002Fgithub.com\u002FSea-Snell\u002FJAX_llama)\n* The JAX\u002FFlax GPT-J and RoBERTa implementation are from [transformers](https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Ftransformers\u002Fmain\u002Fen\u002Findex)\n* Most of the JAX utilities are from [mlxu](https:\u002F\u002Fgithub.com\u002Fyoung-geng\u002Fmlxu)\n* The codebase is heavily inspired by [JAXSeq](https:\u002F\u002Fgithub.com\u002FSea-Snell\u002FJAXSeq)\n","EasyLM 是一个简化大型语言模型（LLM）预训练、微调、评估和服务的一站式解决方案，基于 JAX\u002FFlax 实现。项目的核心功能包括支持 LLaMA、LLaMA 2 和 LLaMA 3 等流行模型的训练与部署，并通过 JAX 的 pjit 功能实现跨多个 TPU\u002FGPU 加速器的大规模并行训练，有效解决单个加速器内存不足的问题。此外，EasyLM 基于 Hugging Face 的 transformers 和 datasets 库构建，提供了易于使用和定制的代码库。该项目适用于需要高效训练和部署大规模语言模型的研究人员和开发者，特别适合在 Google Cloud TPU Pods 上进行多主机训练。","2026-06-11 03:29:53","top_topic"]