# Authored finite-choice and next-token distributions. Original values are in choices.json. system assistant-decision-clarified "Resume checkpoint · I meant the retrieval eval" { version 0.2 theme blueprint direction right option connector_style orthogonal node brief "I meant the retrieval eval" kind input description "Yesterday’s retrieval eval and ranking eval both have saved runs." node retrieval "Retrieval eval · 80%" kind process description "80% · 16/20" node other "Other saved eval · 5%" kind process description "5% · 1/20" node unspecified "Unspecified · 15%" kind process description "15% · 3/20" node rule "App rule" kind decision description "Check saved eval" node result "Resume checkpoint" kind output description "Next · batch 13" node context "Reply context" kind input description "Prompt + app action" node firsttoken "After token one · Resuming" kind process description "Resuming" node nexttoken "After token two · Resuming the" kind output description "Resuming the" edge weigh-retrieval brief -> retrieval edge propose-retrieval retrieval -> rule edge weigh-other brief -> other edge propose-other other -> rule edge weigh-unspecified brief -> unspecified edge propose-unspecified unspecified -> rule edge act rule -> result edge wordone context -> firsttoken edge wordtwo firsttoken -> nexttoken cite "https://docs.typesafe.ai/primitives/choice.md" "TypeSafe Choice · selected option and probabilities" target rule cite "https://huggingface.co/docs/transformers/en/generation_strategies#basic-decoding-methods" "Transformers generation strategies · greedy decoding and sampling" target firsttoken story decision "I meant the retrieval eval" { step read focus brief duration 550 step scores flow weigh-retrieval,weigh-other,weigh-unspecified duration 850 step propose flow propose-retrieval duration 740 step action flow act duration 730 step token-one flow wordone duration 900 step token-two flow wordtwo duration 950 } }