BigScience BLOOM | 3D Parallelism Explained | Large Language Models | ML Coding Series @TheAIEpiphany
BigScience BLOOM | 3D Parallelism Explained | Large Language Models | ML Coding Series  @TheAIEpiphany
Uploaded September 2022 | Updated September 2026, 1 week ago
❀️ Become The AI Epiphany Patreon ❀️
patreon.com/theaiepiphany

πŸ‘¨β€πŸ‘©β€πŸ‘§β€πŸ‘¦ Join our Discord community πŸ‘¨β€πŸ‘©β€πŸ‘§β€πŸ‘¦
discord.gg/peBrCpheKE

In this 4th video of the Large Language Model series I walk you through the BigScience's BLOOM model codebase!

The main focus is on understanding the 3D parallelism:
* Pipeline parallelism
* Model parallelism
* Data parallelism

A set of beautiful engineering ideas that are behind all of the recent scaling efforts and ML success stories!

β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬
βœ… BLOOM code: github.com/bigscience-workshop/Megatron-DeepSpeed
βœ… Ultimate Guide to Scaling Video: youtube.com/watch?v=hc0u4avAkuM
β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬

⌚️ Timetable:
00:00:00 Intro - focusing on the 3D parallelism!
00:02:00 Quick setup
00:05:00 Stepping through the eval script
00:11:13 3D paralellism - model construction
00:15:00 Sharding the embedding table (model parallelism)
00:20:09 Sharding the transformer layer
00:22:30 LayerNorm fused kernels
00:23:50 Sharding the attention layer
00:25:15 ColumnParallel and RowParallel sharding
00:31:30 Synchronizing input and output embedding tables
00:34:45 Building the dataset (data parallelism)
00:39:15 3D parallelism - forward pass
00:39:25 Pipeline parallelism communication
00:43:35 Pass through the sharded embedding table
00:52:15 Pass through the sharded transformer layer
01:01:36 Sharded logit and cross-entropy computation
01:05:30 Recap
01:11:15 Outro

β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬
πŸ’° BECOME A PATREON OF THE AI EPIPHANY ❀️

If these videos, GitHub projects, and blogs help you,
consider helping me out by supporting me on Patreon!

The AI Epiphany - patreon.com/theaiepiphany
One-time donation - paypal.com/paypalme/theaiepiphany

Huge thank you to these AI Epiphany patreons:
Eli Mahler
Petar VeličkoviΔ‡

β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬

πŸ’Ό LinkedIn - linkedin.com/in/aleksagordic
🐦 Twitter - twitter.com/gordic_aleksa
πŸ‘¨β€πŸ‘©β€πŸ‘§β€πŸ‘¦ Discord - discord.gg/peBrCpheKE

πŸ“Ί YouTube - youtube.com/c/TheAIEpiphany
πŸ“š Medium - gordicaleksa.medium.com
πŸ’» GitHub - github.com/gordicaleksa
πŸ“’ AI Newsletter - aiepiphany.substack.com

β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬β–¬

#bloom #bigscience #3dparallelism #scaling
BigScience BLOOM | 3D Parallelism Explained | Large Language Models | ML Coding SeriesOpenAI DALL-E 3 with James Betker (1st author)Day 10: Open NLLB - evaluation data, filtering (Pt 3.)Day 13: Open NLLB - FSDP paper, kicking off the 1st run (Pt 1.)Day 8: Meta NLLB - analyzing the training script, Jais paper (Pt. 1)Channel Update: vacation, leaving Microsoft, approaching 10k subs and more!Neural Descriptor Fields: SE(3)-Equivariant Object Representations for Manipulation Paper ExplainedDay 22: Open NLLB - HBS data analysis, split into Cyrillic & Latin (Pt 3)Day 2: Replicating Metas NLLB - primary & mined dataDo Vision Transformers See Like Convolutional Neural Networks? | Paper ExplainedDay 11: Open NLLB - Apex, mined data, etc.Tanishq Abraham (PhD at 19) - MedARC
Aleksa Gordić - The AI Epiphany |

BigScience BLOOM | 3D Parallelism Explained | Large Language Models | ML Coding Series

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER