{"id":19428898,"url":"https://github.com/eleutherai/improved-t5","last_synced_at":"2026-02-25T08:10:41.626Z","repository":{"id":156319987,"uuid":"615679977","full_name":"EleutherAI/improved-t5","owner":"EleutherAI","description":"Experiments for efforts to train a new and improved t5","archived":false,"fork":false,"pushed_at":"2024-04-15T05:51:48.000Z","size":44731,"stargazers_count":77,"open_issues_count":2,"forks_count":5,"subscribers_count":6,"default_branch":"main","last_synced_at":"2025-04-24T18:47:58.474Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/EleutherAI.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2023-03-18T11:12:12.000Z","updated_at":"2025-01-05T18:09:33.000Z","dependencies_parsed_at":null,"dependency_job_id":"5960e02a-91b6-4493-b9f9-3ccaa119f7d3","html_url":"https://github.com/EleutherAI/improved-t5","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/EleutherAI/improved-t5","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/EleutherAI%2Fimproved-t5","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/EleutherAI%2Fimproved-t5/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/EleutherAI%2Fimproved-t5/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/EleutherAI%2Fimproved-t5/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/EleutherAI","download_url":"https://codeload.github.com/EleutherAI/improved-t5/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/EleutherAI%2Fimproved-t5/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":29815020,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-02-25T05:36:42.804Z","status":"ssl_error","status_checked_at":"2026-02-25T05:36:31.934Z","response_time":61,"last_error":"SSL_read: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-11-10T14:17:06.022Z","updated_at":"2026-02-25T08:10:41.596Z","avatar_url":"https://github.com/EleutherAI.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Improved T5\n\n## Overview\n\n- `configs/`: Configs for pretraining and finetuning\n- `experiments/`: Scripts to launch pretraining and finetuning\n- `convert_weights/`: Scripts to convert T5x checkpoints and upload them to HF\n- `data/`: Seqio scripts for loading datasets\n- `model/`: Gin config for models\n- `tpu-scripts/`: Helper scripts for running training jobs on TPUs\n\n## Running Experiments on TPUs\n\n### Setup\n\nExperiments were run on TPUs. The main scripts involve `send.sh`, `run.sh`, `setup.sh` and `kill.sh` in `tpu-scripts/`. To setup a TPU with the required libraries and dependencies, \n```\nbash send.sh \u003cTPU Node Name\u003e setup.sh\n```\nThen run \n```\nbash run.sh \u003cTPU Node Name\u003e \"bash setup.sh\"\n```\n\nTo run an pretraining/finetuning job,\n```\nbash run.sh \u003cTPU Node Name\u003e \"source env-t5x/bin/activate; cd improved-t5/; bash \u003cscript\u003e\"\n```\n\nIf you need to rerun scripts on the node, you could make sure the node is empty by running\n```\nbash kill.sh \u003cTPU Node Name\u003e\n```\n\nIf you are using a different zone than `us-central2-b`, you will need to changed the `--zone` argument in all the scripts.\n\n## Convert Checkpoints to HF\n\nIn `convert_weights/` you can use scripts to convert T5x checkpoints.\n```\nbash scripts/convert_v1.sh MODEL_SIZE /PATH/TO/T5X_CHECPOINTS/ /PATH/TO/HF_CHECPOINTS/\n```\n\n```\nbash scripts/convert_v2.sh MODEL_SIZE /PATH/TO/T5X_CHECPOINTS/ /PATH/TO/HF_CHECPOINTS/\n```\n\nUse absolute paths [issue](https://github.com/huggingface/transformers/issues/15464#issuecomment-1160318564)\n\nYou can also directly convert T5x checkpoints and upload them to your HF hub. For that see `convert_weights/upload.sh` and `convert_weights/upload-multiple.sh`.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Feleutherai%2Fimproved-t5","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Feleutherai%2Fimproved-t5","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Feleutherai%2Fimproved-t5/lists"}