Skip to content
Snippets Groups Projects

Repository graph

You can move around the graph by using the arrow keys.
Select Git revision
  • 18b5725cacfdfcd6fe14a24b3308580fb445c7ed
  • master default protected
  • add_default_vectorizer_and_pretrained_loading
  • change_system_act_in_env
  • clean_code
  • convlab_exp
  • crossowoz_ddpt
  • e2e-soloist
  • emoUS
  • generalized_action_dicts
  • issue127
  • issue_114
  • nlg-scgpt
  • ppo_num_dialogues
  • pre-training
  • readme
  • remapping_actions
  • robust_masking_feature
  • scgpt_exp
  • soloist
20 results
Created with Raphaël 2.2.015Jul141312865430Jun2928272623222120141310832130May27242317161312111076530Apr29272625242221201918151413111098122Mar211918171510932128Feb262523191817161514102130Jan2624211913731Dec2827222122212220191514109872130Nov2928272413Sep19Aug30Jul718Jun1720May729Apr16215Mar1428Jan29Dec18Nov131Oct30292827262218171513762130Sep2928272524232221201910987542129Aug26241916629Jul27262422181615131097429Jun23181715131025May144229Apr129Febupdate requirements.txtMerge branch 'master' into convlab_expFix training error for domains in tmadd __init__.pyupdate get token loss method for multi-gpu inferencerm reddit dial that have urladd reddit preprocess, update wikidialog preprocess, rm dials that have len(utterance)>256fix slot name for tm2fixupdate configtus unify dataformatRefactoringrm useless funcMerge pull request #74 from ConvLab/convlab_expUpdate requirements.txtadd preprocess.py for opendialkg and 10% wikidialogMerge pull request #73 from ConvLab/convlab_expreplace HF seq2seq trainer to convlab seq2seq trainer to allow more config for prediction such as temperature, do_sampleadd retrieval augmented NLUt5dst multitasknlu multitask expBug fixUpdate DST eval scriptCleanupRefactored setsumbt loss functionssynced github masterMerge pull request #72 from ConvLab/solve_pickling_of_databasesolved by 1) ensuring database exists or is downloaded and 2) import using exec python methodMerge pull request #71 from ConvLab/convlab_expnlg multitask expbugfix, i need to load database manually in vectorbase, otherwise distributed training fails due to not pickle serializable. Also made environment such that it can expect a goal passed to it when resettingBugfix, need to set goal=None as defaultBug fixesupdate setup.py & requirements.txtMerge github masterMerge github_mastermoved plot_results into policy folder and deleted data_loadersMerge remote-tracking branch 'github/master' into github_masterMerge pull request #70 from ConvLab/clean_codeupdate TemplateNLG to accept flatten dialog acts in unified format