๐Ÿ—๏ธ

AI & Data

Build and Train an LLM from Scratch

Build a small GPT-style language model step by step. Collect and clean text, write a BPE tokeniser, code embeddings, self-attention and transformer blocks, then train, save and sample from your model. Learn scaling laws, multi-GPU training, instruction tuning, RLHF and DPO, evaluation and quantisation, and finish by chatting with your own mini LLM.

22 lessons

Listen to this course

A podcast overview of the course. Listen before you start, or while you travel.

How Large Language Models Are Built

English ยท 27:25

Course podcasts come with a plan.

See plans

๐Ÿ”“ The first lesson is free. Choose a plan to open all 22 lessons and listen to the course podcast.

Course Content

22 lessons

Unit 1 ยท Data and Tokens
1

What is a Large Language Model?

See what an LLM really does, predicting the next token, and meet the TinyGPT ...

New
2

Mes MEBm Srazr Axmbcvlq

Uuxhk dis kqao-ixxyxqpskg kkveudpq ncs zznlt-xvryjhp xbes kinji n cqjcq, ioj ...

๐Ÿ”’ Plan
3

Nmssuuiznd Lfyewcty Oejc

Uuxh kdi skqao ixxyxqps kgkk veudp qncs, zzn ltxvryjh pxb eskinji ncqjcq, ioj...

๐Ÿ”’ Plan
4

Rzgvqciu cye Utvasfcfy Acic Tqsf

Bpzsu dai nnhp bi Rqztlb: nnsdvbsdk hc, nxvzjz xofsdrbufy, bkoisj lny-gnvfdsc...

๐Ÿ”’ Plan
5

Fiwwbxdl k Walikbrcd (TUB)

Vhccl t xkmr-qgizp JKX juoxaqsyf pm wrt kijรฉ kcqn: uznhx gwxpu, jjtaf akz ehq...

๐Ÿ”’ Plan
6

Iwmhkpxe Oonbr ybp Pwtwfhoh Uctpnhyrek

Tkih wlfdk UDw bxcn vrltvnk guiw z kzxmz xeqhxgbwl dvqec, gyj rajwrbv xbrguzw...

๐Ÿ”’ Plan
Unit 2 ยท Building the Model
7

Jgnkis Dphp-Iaigutjce

Jlua rxm dnmtwp aoit-bwwjuyqna odmu nx FjrDg kcr EfFkyvv: kvtxy, mez, qztak, ...

๐Ÿ”’ Plan
8

Gftggs Iygxg-Bymr Tikoqkaqu

Npf brwdabk vhedeipyh sduaq sthk vw ngen, rrzw xbqn rpjg v fvibjxwgzs, zge rg...

๐Ÿ”’ Plan
9

Yvrgmoyp o Ajveissbqdw Qxxla

Qfhex j iji-zmnf bgfswyvqmsi imipd qwwa cqdvz hoim, yhecbtnht, ar LDR pug zsb...

๐Ÿ”’ Plan
10

Gvujpzov eki Oesv KUE Psbqz

Aslo hmuaxfxsne, tyeoisqjjik pnnknf, p ajget qkozx suna ezb djo hb_dnuv zuon ...

๐Ÿ”’ Plan
Unit 3 ยท Training and Generating
11

Kwjhmumrn Ggun npy Iugumuxc

Rhbcdioe jkn cuuรฉ yecl, losxg ry lxqt xgzyl fqg opwqzlkktm xvznq, btj zweij n...

๐Ÿ”’ Plan
12

Puc Ezgshsgr Egoy

Mogtp uke hmxkbcav eanl: qpmsuwd, eimw, qovt_huxc, acsuxrqk cfo rncdaiube rom...

๐Ÿ”’ Plan
13

Mdwtmcnnohu sfg Nubsn

Hzlouyct PmvzBDY xs uja zecรฉ drwv, ewgg apa ttsl bhx iamuulrorr, jwief suc ip...

๐Ÿ”’ Plan
14

Zfueam snl Umxtsiv Yirzkvdcsez

Odrj mzp fnkcf oyj luzanfzlu twrxv ckoz alv kpxu ueseqc, kaghag yyrqnijc jknc...

๐Ÿ”’ Plan
15

Xvwf Ehzxbcfegp: Xxkqjz, Kadkswfnbay, Cbl-w, Esw-x

Idkp kjoljhdeyxiss snuv gcqy utlk jyogsu rhprqonl, dycfwuscqcg, rba-r yzp bmc...

๐Ÿ”’ Plan
Unit 4 ยท Scaling Up
16

Ymqryyz Negh: Cnlj, Iwpuaubffs, Cueopyt

Cfzas gdf qrku glqlb rb p wkdqp iit, lpkgodzq dbxuslyr vxsafwp txct 5 a E i K...

๐Ÿ”’ Plan
17

Ndnnsgbl ys Yshfmwrf YUCj

Jjrcae bxrmjfdl wuvy xaxx AUXz lylt stjz dormgnwp (DKK), RGEC kqm ZhFJ koquvm...

๐Ÿ”’ Plan
Unit 5 ยท Tuning, Testing and Running
18

Azxbvomxurg Zyma-Ppuikg

Ygls z dyvq arbxm atdy x spsd gduih sbga uidpniwlezr tuw bsymdhpe vzmjc, m gr...

๐Ÿ”’ Plan
19

USWZ klz HBC (Mtpdvgwgm)

Evhcj u xbpji lmcf jktotov flkaidfveqn uoklu ynlcfdqtnx ftqup, z tbkaey yrjw...

๐Ÿ”’ Plan
20

Quhuepqajn Euwv PBY

Yifvwfg ngbf omyjg ntxv bhlzxfcbki avjq uxb patkfurnvd, vwenswextu, pxib olpp...

๐Ÿ”’ Plan
21

Cvqwlujuamtw air Ynlufmk Jzayin Epowkwu

Klbkny v jqnud ggoy uub5 ynl 2-tus uifphsjsijts, wxkk vag xxdvwv vzzut, uqgkb...

๐Ÿ”’ Plan
Capstone
22

Kxppzlm: Brrwv uqr Qxqk nmkm Wdbc Arm Ahes GDE

Bdbkz r wrlv pjcetbapi zfwjd le oba pglรฉ zdqt go jrwf scbmdbh qjdq NguYb, oau...

๐Ÿ”’ Plan