Skip to content

implementation of the self attention architecture in a decoder only auto regressive nlp model trained on shakespeare's work

Notifications You must be signed in to change notification settings

nischalj10/gpt2

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

building gpt2 - starting with training on shakespere's work. char level autoregressive model.

after 24 hrs of training gpt.py on M2 Air lol -

step 0: train loss 4.2846, val loss 4.2820

step 500: train loss 1.6359, val loss 1.8073

step 1000: train loss 1.3473, val loss 1.5734

step 1500: train loss 1.2373, val loss 1.5051

step 2000: train loss 1.1626, val loss 1.4828

step 2500: train loss 1.0949, val loss 1.4830

step 3000: train loss 1.0391, val loss 1.4924

step 3500: train loss 0.9833, val loss 1.5162

step 4000: train loss 0.9287, val loss 1.5288

step 4500: train loss 0.8720, val loss 1.5567

0.9912359714508057

the holy 3000 tokens of shakespere wisdom

ESTA: Tell hie, my grave before thy sin son: If I belike soft in Edward, should be found, I do before thee burnide the woe.

Servant: Ay, she as her father, she is married; When she hath pursued him in my feet fear To slaughter you all his hatred by.

STER: I wrong to the tribunes, fatally goes, The which height and Bolingbrook.

BUSHY: She it makes hences from some and nature; Or never will be govern'd with my occupy: Good night to the citing of your delight, And wo to meet That I'll procure your life, With poltrous, and known themselve of them. If they should love both take away craft, And bid the time light, so strange sumility, And that make up their tender delosures From their robbing face hath past of will, And then my opposite will he humble be with powerful. Whom grieve that so fair duty in this lives So make a drown of welcomend; as that am no passing Is true to possesses that uncourage is, go a rencourage tender, a voison, the shadowment, which has bearing it: you may be praised, hanging, now all fitter and the precious mistress, which you call the jest fingers, and unnaturate in the dalm. Come you and son: I wot his uncertainer, both your hand.

First Lord: Bring it the father of our Cretandness,--

As your made us.

ESCALUS: Shepherd a pleasant senators o'er the death, to dream the offender and ospect: what a presset act act so mock it lig, and please but it becut it off, it inclines.

BHORSELORD: What, to-morrow? close it is true.

POMPEY: Prayers sit now, sir, quarrel?

HERMIONE: The breast will I leave to the leave o' twention; be married, Jove of Py; you that shall been such at homily server, he shall pass all till has he was worn.

First Citizen: It is your as is done: in revenge one here tellings, you my word.

Citizens: I condign two one present yew-time, one proud, with respection; so gratty increaming.

CATESS OF AUMPREa: Why, let's fight and Somervilege, wholesome nature Our gracious wings and perish-line full out Such windeeds where I stables juster. have let you down!

BRAKENBURY: Paganet! Why, then, shall you take Henry from Bolingbroke?

RICHARD: Yes, are you conforce to boot a strayer?

ELBOW: I was one confessed his honour frongs with me alone.

LORDS: It must be so: farewell arrest you sow to senthing you wise your graces. For your beewark this fast perial title, you see how to be hewn, not a second down! If the doubt of foot, let it but every bleed, And say, it will bring a doubtful fear, Environed with reverin sprown: there Cry young Warwick is here to gentlemen, England us to the Lord Bone, of that news.

GEOl CAstle: Harry of Greece we pray, we do go so, And tell our father, opinions, how have wet letters? if He we be rod,--

JULIET: At one time shall hath made me go by me conceit in these trifles And by through she bsteeds would be cloudegly That all he acts like disloyal, as at if Hnnerible of sestern: is it so?

RIVERS: His goodly sits can would sit for the realm, With cloud off thither Troy.

DUKE ERVINCENTIO:

commemorating this with -

alt text

About

implementation of the self attention architecture in a decoder only auto regressive nlp model trained on shakespeare's work

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published