# machine-learning
- b
BrokeSkill
10/15/2022, 7:05 PMok yeah what exists that has a good premodel and can be trained - h
hecko
10/15/2022, 7:06 PMi don't think those exist - h
hecko
10/15/2022, 7:06 PMbecause goodness is mostly correlated with size - h
hecko
10/15/2022, 7:07 PMand if it's too big it won't fit on colab - b
BrokeSkill
10/15/2022, 7:07 PMok a normal premodel that can be trained - h
hecko
10/15/2022, 7:08 PMdialogpt can be trained i think - h
hecko
10/15/2022, 7:08 PMbut don't train it on reddit comments because that's what it was already trained on - b
BrokeSkill
10/15/2022, 7:08 PMwhat else then - b
BrokeSkill
10/15/2022, 7:09 PM._. - b
BrokeSkill
10/15/2022, 7:10 PMi think hes good - b
BrokeSkill
10/15/2022, 7:10 PMim not sure tho - b
BrokeSkill
10/15/2022, 7:11 PMmakes sense - b
BrokeSkill
10/15/2022, 7:12 PMbecause its mad dumb - b
BrokeSkill
10/15/2022, 7:12 PMmaybe conversations - b
BrokeSkill
10/15/2022, 7:12 PMor questions - b
BrokeSkill
10/15/2022, 7:13 PMafter like two questions it runs in a loop - h
hecko
10/15/2022, 7:18 PMmight be due to a low temperature setting - b
BrokeSkill
10/15/2022, 7:21 PMhow do i change it - b
BrokeSkill
10/15/2022, 7:23 PM? - b
BrokeSkill
10/15/2022, 7:28 PMyeah can i use a dataset ? WHich isnt a whole terrabyte and is good - h
hecko
10/15/2022, 7:38 PMare you looking for a dataset or - b
BrokeSkill
10/15/2022, 7:42 PMYeah for a chatbot - b
BrokeSkill
10/15/2022, 7:42 PMThat isnt big - b
BrokeSkill
10/15/2022, 7:42 PMBut good - h
hecko
10/15/2022, 7:43 PMwell i guess there's this https://www.cs.cornell.edu/~cristian/Cornell_Movie-Dialogs_Corpus.html - b
BrokeSkill
10/15/2022, 7:45 PMSo which of them is better 1 month of 2022 reddit Training data or cornell movie Dialogs - h
hecko
10/15/2022, 8:05 PMdialogpt is already trained on reddit - h
hecko
10/15/2022, 8:06 PMand it really depends on what flavor of chatbot you want - b
BrokeSkill
10/15/2022, 8:28 PMConversational and can answer questions - b
BrokeSkill
10/15/2022, 9:07 PMAnd its dumb so nah