konabuta / my-scratch-book

MIT License
1 stars 0 forks source link

Blog: Techniques for training large neural networks #14

Open konabuta opened 8 months ago

konabuta commented 8 months ago

Blog: Techniques for training large neural networks

Link: https://openai.com/research/techniques-for-training-large-neural-networks

This blog explains the parallism technologies for training large neural networks. image

konabuta commented 8 months ago

Data parallelistm

  1. indendently compte the gradient on each worker.
  2. average the gradents across workers.
  3. independently compute the same new parameters on each worker.

Step can be bottleneck becuase it requires transferring data.

konabuta commented 8 months ago

Pipeline parallelism

konabuta commented 8 months ago

Tensor parallelism

konabuta commented 8 months ago

Mixture-of-Experts (MoE)

konabuta commented 8 months ago

Others

konabuta commented 8 months ago

Reference