AI-212 AI & Data

Ray & Large-Scale Model Training

Engineer the training platform itself: Ray as the distributed substrate, DeepSpeed ZeRO and FSDP as the sharding layer, and Kubernetes as the scheduler underneath. Learners stand up a multi-node training job that survives node loss, tune it for utilisation rather than raw speed, and defend a run plan with cost, scaling-law and stop-criteria evidence rather than intuition.

B5Advanced

6modules

42skill atoms

1role journey

Curriculum

What this course covers.

6 modules, 42 named skill atoms. Expand any module to see them.

1Ray Core Concepts7 skill atoms

tasks and actorsobject store and zero-copy refsplacement groups for gang placementGPU resource requestsRay Dashboard inspectiondriver vs worker failure semanticsserialisation pitfalls on large objects

2Ray Train and Ray Tune7 skill atoms

TorchTrainer and ScalingConfigworker groups and per-worker setupcheckpoint reporting into the runASHA and population-based training schedulerssearch space definitiontrial concurrency vs GPU supplyearly stopping on a plateau

3DeepSpeed ZeRO and FSDP7 skill atoms

ZeRO stages 1 2 and 3optimizer state and parameter shardingCPU and NVMe offload costFSDP auto-wrap policiesfull vs sharded state dictmixed precision policy choiceZeRO vs FSDP selection criteria

4Multi-Node Scheduling on Kubernetes7 skill atoms

KubeRay operator and RayCluster CRDhead and worker node poolsgang scheduling and pending-pod deadlockGPU device plugin and node taintsshared storage for datasets and checkpointsNCCL and network env tuningspot interruption handling

5Fault Tolerance and Elastic Training7 skill atoms

worker failure retry policyresume from last durable checkpointelastic world-size changespreemption drain hookshung-rank detection and job killidempotent data sharding on restartblast radius of one bad node

6Cost, Utilisation and Scaling Laws7 skill atoms

GPU-hour cost per runMFU as the utilisation metriccompute-optimal sizing as a floor not a targetinference-cost-aware over-training past Chinchilla ratiostotal cost of ownership over the serving lifereserved vs spot capacity mixstop criteria and diminishing returns

Where it fits

AI-212 in a role journey.

This course appears in 1 of our 45 role journeys. Here is what a learner takes immediately before and after it in each.

AI Infrastructure Engineer

Capstone stage

CL-405→AI-212→journey complete

Roles this course serves

AI Infrastructure Engineer

The capability ladder

This course is authored to band B5.

Every course we run is written to one rung of the CASI ladder, so a plan can be assembled to take a team from where they are to where they need to be.

What do B1–B6 mean?The CASI Capability Ladder — click to expand

Every course targets a band on the CASI Capability Ladder — our six-band proficiency scale, anchored to open standards (O*NET, ESCO, NICE, NIST AI RMF, Bloom's). A band tells you how deep a course goes, and what evidence proves it.

What the learner can doTypical evidence

AwareUnderstands concepts and vocabulary; uses tools with guidance

Knowledge checks

FoundationPerforms standard tasks correctly in familiar contexts

Guided labs, autograded exercises

PractitionerDelivers complete pieces of work independently

Scenario labs, proctored hands-on exams

ProfessionalHandles production-grade complexity, trade-offs and failure modes

Break-fix drills, design defenses

AdvancedEngineers systems end-to-end under constraints; leads others

Rubric-scored capstones, vivas

ExpertSets direction; recognised authority across teams

Portfolio + panel evaluation

A note on B6. Courses in this catalog target B1–B5. B6 is not taught — it is recognised, through a portfolio and a panel, once someone is setting direction for others. Every journey here is built to land a learner at B5.

Other AI courses at this level.

AI-208

Run AI-212 for your team.

This course runs at several lengths depending on how deep you need to go and how much of it your people already have. Tell us who is being trained and we will scope it.

Add it to a training plan Talk to our team Check your team’s level free

Ray & Large-Scale Model Training

What this course covers.

AI-212 in a role journey.

AI Infrastructure Engineer

Roles this course serves

This course is authored to band B5.

Other AI courses at this level.

Model Compression: Quantization, PEFT, QLoRA & Distillation

GPU Engineering for AI (CUDA · cuDNN · TensorRT)

Preference Tuning: RLHF, DPO & Reward Models

Knowledge Graphs & GraphRAG

Multi-Agent Systems & Orchestration

LLM Inference Optimization & Serving (vLLM · LiteLLM)

Run AI-212 for your team.