Slide 1

Slide 1 text

Ray in 2023 Robert Nishihara

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

12x 50% 40% 10x 5x 30% Why Ray? faster cheaper cheaper cheaper faster cheaper

Slide 4

Slide 4 text

As AI capabilities have grown, so have the challenges Scale Future readiness Cost These are the challenges Ray was built for

Slide 5

Slide 5 text

Anyscale Endpoints - fine-tuning Llama-2-7B GPT-4 fine-tuned 86% 3% 78% Superior task-specific performance at 1/300th the cost of GPT-4!

Slide 6

Slide 6 text

Spark SageMaker $0 $20 $40 $60 $3.5 $7.3 $57 AWS Cost to process 1M images $2.5 Batch inference - costs

Slide 7

Slide 7 text

Anyscale Endpoints Cost efficient LLM inference Anyscale Endpoints Single GPU optimizations Multi-GPU modeling Inference server Autoscaling Multi-region, multi-cloud $1 / million tokens (Llama-2 70B)

Slide 8

Slide 8 text

No content

Slide 9

Slide 9 text

No content

Slide 10

Slide 10 text

No content