You can move around the graph by using the arrow keys.
Created with Raphaël 2.2.010Feb9876432131Jan272625242321201918171614964131Dec302928272523222019161514121197652130Nov2926252423222119181716141110989878765432131Oct302827262524212019181716151413121110765432129Sep2827232223212017161514131210976231Aug30292322201817151211109876532131Jul30292827262726252625lint updateload-checkpoint…load-checkpoint-on-more-gpusfix importslint updateinitial commitmore cleanup, add reset_for_finetuning back inpeter/rewrite_l…peter/rewrite_load_checkpointadd some debugging and load local checkpointsSimplify inheritance layers (#639)mainmainremove zero_sharding code (#645)Community: Add the CTranslate2 integration (#644)Freeze positions and report more fine-grained lossmarcin/pos_freezemarcin/pos_freezefixngoyal_continue…ngoyal_continue_training_on_larger_gpusMerge branch 'main' into reptition_and_factual_nucleusreptition_and_f…reptition_and_factual_nucleusfixfix for NFSinitial implementation of cm3/fim objectivecm3_oscm3_osremove TransformerDecoderLayerSimplify-layers…Simplify-layers-of-inheritancelint updateadd back TransformerLanguageModel for compatibilitytry update the test_sequence_parallel.pyremove TransformerDecodertwo paths for local vs nfs checkpointsremove LanguageModelUnify-transformer_lm_megatron-and-transformer_lm (#634)api configs for modelsopt_instruct_re…opt_instruct_rebasedUpdate CODEOWNERS (#635)update codeowners (#631)Merging model_parallel with models (#626)Merge branch 'opt_instruct_rebased' of github.com:facebookresearch/metaseq into opt_instruct_rebasedAdd max gen tokenschangesInitialization changes (#579)susan/pos_init_…susan/pos_init_on_bigbigReplace optiml_paper_v1.pdf with placeholder (#623)fixesupdate codeowners (#624)Fix dataset errorMerge branch 'main' into peter/rewrite_load_checkpointadd binhs small changespeter/api_exportpeter/api_exportMerge remote-tracking branch 'origin/reshard-fsdp' into peter/api_export.streamandbatchstreamandbatchFixes to make data loading work